Transaction risk identification method, device and storage medium

Abstract

The specification provides a transaction risk identification method, device and storage medium. In the method, in response to a risk identification request of a target transaction, transaction information of the target transaction is obtained, a target question set is determined based on the transaction information by a first large model, and then the target transaction is risk identified based on the target question set. The target question set includes at least one guide question for risk identification of the target transaction. The first large model is trained to determine different question sets for different transactions, and the question set determined by the large model is the minimum question set that can achieve the risk identification target.

Description

[0001] This application claims priority to Singapore Patent Application No. 10202500139Y, filed on January 16, 2025, the entire contents of which are incorporated herein by reference. Technical Field

[0002] This specification relates to the field of Internet technology, and in particular to a method, device and storage medium for identifying risks in transactions. Background Technology

[0003] With the development of the internet and payment technology, payment scenarios are becoming increasingly diversified, and the risks threatening payment platforms and users (such as fraud) are also gradually increasing. Therefore, payment platforms are placing greater emphasis on transaction risk control. For example, in international transaction scenarios, when a user initiates a payment request for a transaction, the payment platform requires the user to upload supporting documents related to the transaction and generates a risk identification task for that transaction. When performing the risk identification task, the payment platform's operators need to manually review the aforementioned supporting documents to determine whether the transaction is risky.

[0004] When operators identify risks for different transactions, the key identification points they need to consider, based on the supporting documentation, vary. Therefore, operators need to memorize different identification points for different types of transactions, requiring a high level of risk control experience. As the number of transactions increases, some risk identification tasks may be assigned to less experienced operators. These operators may need to temporarily refer to the operation procedure manual, leading to longer risk identification times, impacting the timeliness of transaction payments, and consequently affecting user experience.

[0005] The information in the background section is merely information known only to the inventor and does not imply that such information had entered the public domain before the date of this application, nor does it imply that it can be considered prior art in this disclosure. Summary of the Invention

[0006] This specification provides a method, device, and storage medium for risk identification in transactions. By using a first major model to determine the guiding problem for risk identification, different transactions can have different guiding problems, which is more flexible and targeted. This can reduce redundant guiding problems, shorten the risk identification time of transactions, and lower the learning threshold for operators.

[0007] In a first aspect, this specification provides a method for risk identification of a transaction, comprising: in response to a risk identification request for a target transaction, obtaining transaction information of the target transaction; determining a target question set based on the transaction information of the target transaction using a first master model, the target question set including at least one guiding question for risk identification of the target transaction, wherein the first master model is trained to determine different question sets for different transactions, and the question set is a minimum question set capable of achieving the risk identification objective; and performing risk identification of the target transaction based on the target question set.

[0008] Secondly, this specification also provides an electronic device, comprising: at least one storage medium storing at least one instruction set for risk identification of a transaction; and at least one processor communicatively connected to the at least one storage medium, wherein the at least one processor reads the at least one instruction set during operation and, according to the instructions of the at least one instruction set: in response to a risk identification request for a target transaction, obtains transaction information of the target transaction; determines a target problem set based on the transaction information of the target transaction using a first large model, the target problem set including at least one guiding question for risk identification of the target transaction, wherein the first large model is trained to determine different problem sets for different transactions, and the problem set is a minimum problem set capable of achieving the risk identification objective; and performs risk identification of the target transaction based on the target problem set.

[0009] Thirdly, this specification provides a computer-readable non-transitory storage medium, wherein the computer-readable non-transitory storage medium stores at least one instruction set, which, when executed by at least one processor, implements the transaction risk identification method as described in the first aspect.

[0010] Other functions of the transaction risk identification methods, devices, and storage media provided in this specification will be partially listed in the following description. The figures and examples described below will be readily apparent to those skilled in the art. The inventive aspects of the transaction risk identification methods, devices, and storage media provided in this specification can be fully understood through practice or use of the methods, apparatus, and combinations described in the detailed examples below. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A schematic diagram of an application scenario for risk identification of transactions provided according to an embodiment of this specification is shown; Figure 2 A hardware structure diagram of an electronic device provided according to an embodiment of this specification is shown; Figure 3 A flowchart of a risk identification method for a transaction provided according to an embodiment of this specification is shown; Figure 4 A flowchart of a training method for a first large model provided according to an embodiment of this specification is shown; Figure 5 A schematic diagram illustrating a process for risk identification of transactions according to an embodiment of this specification is shown. Figure 6 A schematic diagram illustrating another process for risk identification of transactions provided according to an embodiment of this specification is shown; Figure 7 A schematic diagram illustrating yet another process for risk identification of transactions provided according to an embodiment of this specification is shown; Figure 8 A schematic diagram illustrating yet another process for risk identification of transactions according to embodiments of this specification is shown; and Figure 9 A flowchart of a training method for a second large model provided according to an embodiment of this specification is shown. Detailed Implementation

[0013] The following description provides specific application scenarios and requirements for this specification, intended to enable those skilled in the art to make and use the contents of this specification. Various partial modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles defined herein can be applied to other embodiments and applications without departing from the spirit and scope of this specification. Therefore, this specification is not limited to the embodiments shown, but rather to the widest scope consistent with the claims.

[0014] The terminology used herein is for the purpose of describing particular exemplary embodiments only and is not restrictive. For example, unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “the” used herein may also include the plural forms. When used in this specification, the terms “comprising,” “including,” and / or “containing” mean that the associated integers, steps, operations, elements, and / or components are present, but do not exclude the presence of one or more other features, integers, steps, operations, elements, components, and / or groups, or that other features, integers, steps, operations, elements, components, and / or groups may be added to the system / method.

[0015] Considering the following description, these and other features of this specification, as well as the operation and function of the related components of the structure, and the economy of assembly and manufacture of the parts, can be significantly improved. All of these form part of this specification with reference to the accompanying drawings. However, it should be clearly understood that the drawings are for illustrative and descriptive purposes only and are not intended to limit the scope of this specification. It should also be understood that the drawings are not drawn to scale.

[0016] The flowcharts used in this specification illustrate operations implemented according to some embodiments of this specification. It should be clearly understood that the operations in the flowcharts may not be implemented in a sequential order. Instead, the operations may be implemented in reverse order or simultaneously. Furthermore, one or more additional operations may be added to the flowcharts. One or more operations may be removed from the flowcharts.

[0017] Typically, after a user makes a payment for a target transaction using a terminal device, the terminal device sends the payment to the payment platform's service device (e.g., Figure 1 The server (300) sends a payment request. Upon receiving the payment request, the service device generates a risk identification task corresponding to the payment request and adds the risk identification task to the task queue. It can be understood that the service device can maintain an ordered task queue according to the order of payment requests. The service device can also assign each risk identification task in the task queue to different operators. After receiving the risk identification task corresponding to the target transaction, the operator can trigger risk identification for the target transaction.

[0018] Before identifying a target transaction, operators can obtain the transaction's payment information and supporting documentation submitted by the user. Understandably, the supporting documentation required from the payment platform will vary depending on the transaction. For example, if the payment platform is a cross-border platform and the target transaction is under general trade, the platform may require the user to upload the transaction contract, invoice, and logistics documentation (such as shipping documents or delivery orders). However, if the target transaction is under Ex Works (EXW), the platform may require the user to upload the transaction contract and invoice. This is because under EXW trade, the buyer picks up the goods themselves, and therefore the user cannot provide proof of shipment logistics.

[0019] The payment platform may remind users to upload supporting documents before they make a payment; or it may remind users to upload supporting documents after they make a payment; or, if the operator finds that some supporting documents are missing during the risk identification process of the target transaction, the payment platform may notify the user to submit the missing supporting documents. This manual does not restrict the timing of uploading supporting documents.

[0020] There are multiple ways for operators to perform risk identification on a target transaction, and two of them are exemplified below for illustration.

[0021] Mode 1: Manual identification Experienced operators will develop a set of SOP (Standard Operating Procedure) to train inexperienced junior operators. The SOP includes identification points for different transactions, and preset rules corresponding to each identification point. During the identification process, once some preset rules are determined to be hit, the risky transaction can be identified, while some preset rules need to be combined to identify the risky transaction.

[0022] For example, the identification point is: Has the certification material been tampered with? The preset rule corresponding to this identification point is: If the certification material has been tampered with, the current transaction is a risky transaction; or, if the certification material has not been tampered with, the current transaction may not be a risky transaction, and judgment shall be made in combination with other identification points. If the operator determines that part of the certification materials of the target transaction has traces of tampering, it is determined that the target transaction is a risky transaction.

[0023] For example, identification point #1 is: Is the establishment time of the seller associated with the target transaction more than X days? The preset rule corresponding to identification point #1 is: If the establishment time of the seller is more than X days, the current transaction has no risk in the dimension of identification point #1; if the establishment time of the seller is less than X days, the current transaction may be a risky transaction. Identification point #2 is: Is the shipper in the certification material consistent with the payee in the payment information of the target transaction? The preset rule corresponding to identification point #2 is: If the shipper and the payee are consistent, the current transaction has no risk in the dimension of identification point #2; if the shipper and the payee are inconsistent, the current transaction may be a risky transaction. Identification point #3 is: Is the consignee in the certification material consistent with the payer in the payment information of the target transaction? The preset rule corresponding to identification point #3 is: If the consignee and the payer are consistent, the current transaction has no risk in the dimension of identification point #3; if the consignee and the payer are inconsistent, the current transaction may be a risky transaction.

[0024] If the operator determines that the seller was established only 2 days before the target transaction is initiated, and 2 < X, at this time, it cannot be directly determined that the target transaction is a risky transaction based on the preset rule corresponding to identification point #1. The operator needs to combine the preset rules corresponding to identification point #2 and identification point #3. If the shipper and the payee are inconsistent, or the consignee and the payer are inconsistent, it can be determined that the target transaction is a risky transaction at this time.

[0025] After receiving the identification task, the operator performs risk identification on the target transaction according to the trained SOP. For example, the operator can determine the transaction pattern of the target transaction through the supporting documents uploaded by the user and the payment information obtained in the payment request, and then perform risk identification on the target transaction based on all the identification points corresponding to the transaction pattern, thereby determining whether the target transaction is a risky transaction.

[0026] If the target transaction is a risky transaction, the payment platform will display a prompt message to the user through the terminal device, indicating that the target transaction has been identified as a risky transaction and payment cannot continue, or directly indicating that the payment has failed. If the target transaction is not a risky transaction, the payment platform will display a prompt message to the user through the terminal device, indicating that the target transaction is not a risky transaction or that the risk identification of the target transaction has passed, and prompting the user to continue to the next payment operation.

[0027] Method 1, which provides risk identification, relies heavily on operator experience and requires operators to memorize Standard Operating Procedures (SOPs). Due to the length of these SOPs, accurate memorization is difficult. If operators need to refer to the SOPs during risk identification, it is time-consuming, resulting in poor timeliness. As transaction volume increases, the number of risk identification tasks becomes overwhelming, leading to a backlog and further compromising timeliness, impacting user experience. Furthermore, some operators may overlook key identification points, preventing the payment platform from accurately identifying risky transactions.

[0028] Method 2: Guided Recognition All identification points in the aforementioned SOP are transformed into multiple guiding questions, which are then embedded into the payment platform. For example, when a user performs a payment operation for a target transaction, the payment platform generates a risk identification task corresponding to the target transaction. When the payment platform detects a risk identification request triggered by an operator for the target transaction, it displays the multiple guiding questions to the operator. The specific form of the guiding questions is, for example, the identification points described above, and will not be repeated here. For each guiding question, the operator can fill it out based on the supporting documents uploaded by the user and the obtained payment information. After the operator fills out all the displayed guiding questions, they will have a comprehensive understanding of the target transaction, thereby determining the risk identification result of the target transaction.

[0029] Once the operator has determined the risk identification results for the target transaction, the operation can be referred to the description in Method 1, and will not be repeated here.

[0030] Method 2, which provides a risk identification approach, transforms all identification points in the Standard Operating Procedure (SOP) into guiding questions. This guides operators to utilize all identification points to identify risks in the target transaction, addressing the issue of operators missing key identification points. However, this method presents the same guiding questions to operators across different transactions, lacking flexibility, and some identification points are not applicable to all transactions. For example, for general trade transactions, logistical information is not required for risk identification. Therefore, this method of using all identification points for transaction risk identification results in a longer risk identification time, impacting user experience.

[0031] In view of this, this specification provides a method, device, and storage medium for risk identification in transactions. It can determine a set of target questions for risk identification based on the transaction information of the target transaction, and can determine different sets of target questions for different transactions, offering flexibility and specificity. Furthermore, the number of guiding questions included in the target question set is the minimum required to achieve the risk identification objective, thereby reducing the time required for risk identification and improving the user experience.

[0032] The following is combined Figure 1 This manual introduces the application scenarios provided.

[0033] Figure 1 This diagram illustrates an application scenario for risk identification in transactions, based on an embodiment of this specification. Figure 1 As shown, scenario 001 may include: user 100A, operator 100B, terminal device 200, server 300, risk transaction identification device 400 (hereinafter referred to as identification device 400), and training device 500.

[0034] Figure 1 The example application scenario includes three stages: the training stage, the payment stage, and the risk identification stage. These three stages will be described in detail below.

[0035] (a) Training Phase During the training phase, training equipment 500 was used to train the first and second largest models.

[0036] Server 300 can communicate with terminal device 200, receiving data from or sending data to terminal device 200. Server 300 can provide various services; for example, server 300 is a backend server providing payment services to terminal device 200.

[0037] For example, the training device 500 can obtain transaction information of historical transactions and the corresponding set of historical questions from the server 300. This set of historical questions can be the minimum set of questions that experienced operators can determine for each historical transaction to achieve the risk identification objective. The training device 500 can generate a first training sample based on the transaction information of a historical transaction and the corresponding set of historical questions. Similarly, the training device 500 generates multiple first training samples. Furthermore, the training device 500 can train a first large model based on multiple first training samples. The first large model can determine different sets of questions for different transactions, and the determined set of questions is the minimum set of questions that can achieve the risk identification objective. The specific training process of the first large model is described in detail in the following embodiments.

[0038] For example, the training device 500 can also obtain the input information corresponding to the historical question set from the server 300. This input information is filled in by the operator during the risk identification process of historical transactions, and can be understood as the operator's answers to each question in the historical question set. The training device 500 can also obtain the risk labeling information corresponding to the aforementioned historical transactions from the server 300. This risk labeling information can be the operator's risk identification result for that historical transaction. The training device 500 can generate a second training sample based on the input information and risk labeling information corresponding to a historical transaction. Similarly, the training device 500 can generate multiple second training samples. Furthermore, the training device 500 can train a second large model based on multiple second training samples. The second large model has the ability to identify risks in transactions. The specific training process of the second large model is described in detail in the following embodiments.

[0039] The first and second trained models can be deployed to server 300. Alternatively, the first trained model can be deployed to server 300, and the second trained model can be deployed to recognition device 400. Or, both the first and second trained models can be deployed to recognition device 400. Of course, there are other deployment methods, which will not be listed here.

[0040] Training device 500 is an electronic device with certain data processing capabilities. Training device 500 can execute the data and instructions of the training method for the first or second large model described in this specification, and can execute or be used to execute said data and instructions. For example, training device 500 may include hardware devices with data information processing functions and the necessary programs required to drive the hardware devices to operate.

[0041] (ii) Payment stage During the payment phase, user 100A initiates a payment request for the target transaction via terminal device 200. In response to this payment request, server 300 can generate a risk identification task corresponding to the target transaction. In some embodiments, server 300 can maintain a task queue and add the risk identification task corresponding to the target transaction to the task queue.

[0042] For example, user 100A can execute related events through terminal device 200. For instance, terminal device 200 may have clients installed on one or more service platforms, which provide user 100A with the ability to execute related events. These service platforms include, but are not limited to: web browser platforms, search platforms, chat platforms, shopping platforms, service platforms, video platforms, financial management platforms, payment platforms, etc.

[0043] Terminal device 200 can present an interactive interface to user 100A, allowing user 100A to perform certain events through this interface. For example, in response to user 100A's operation, terminal device 200 displays the interactive interface of a payment platform to user 100A. User 100A can trigger a payment process for a target transaction through the payment platform's interactive interface. Terminal device 200 sends a payment request corresponding to the target transaction to server 300, and server 300 responds to the payment request by generating a risk identification task corresponding to the target transaction.

[0044] Terminal device 200 may include mobile devices, tablets, laptops, built-in devices in motor vehicles, or similar content, or any combination thereof. In some embodiments, the mobile device may include smart home devices, smart mobile devices, virtual reality devices, augmented reality devices, or similar devices, or any combination thereof. In some embodiments, the smart home device may include smart TVs, desktop computers, etc., or any combination thereof. In some embodiments, the smart mobile device may include smartphones, personal digital assistants, gaming devices, navigation devices, etc., or any combination thereof. In some embodiments, the virtual reality device or augmented reality device may include virtual reality headsets, virtual reality glasses, virtual reality patches, augmented reality headsets, augmented reality glasses, augmented reality patches, or similar content, or any combination thereof. For example, the virtual reality device or the augmented reality device may include AR glasses, head-mounted displays, VR, etc. In some embodiments, the built-in device in the motor vehicle may include an in-vehicle computer, an in-vehicle television, etc. In some embodiments, terminal device 200 may be a device with positioning technology for locating the position of terminal device 200.

[0045] (III) Risk Identification Phase During the risk identification phase, server 300 can distribute risk identification tasks from its task queue to identification device 400, or identification device 400 can request one or more risk identification tasks from server 300. After identification device 400 obtains one or more risk identification tasks, the operator can trigger the execution of a specific risk identification task. The following explanation uses the example of an operator triggering the execution of a risk identification task corresponding to a target transaction.

[0046] The identification device 400 is equipped with a pre-trained first major model. When operator 100B triggers a risk identification request for a target transaction, the identification device 400 obtains the transaction information of the target transaction and determines a target question set based on this information using the first major model. The target question set includes at least one guiding question for risk identification of the target transaction. Subsequently, the identification device 400 can perform risk identification on the target transaction based on the guiding question in the target question set. During the risk identification process, a second major model may or may not be used.

[0047] For example, instead of using the second major model, the operator can make a comprehensive judgment on the target transaction based on the guiding questions in the target question set to obtain a risk identification result, and then input the risk identification result into the identification device 400. Alternatively, the identification device 400 can perform risk identification on the target transaction based on each guiding question in the target question set, thereby obtaining a risk identification result.

[0048] For example, a second major model is used, and this second major model is deployed on identification device 400. Identification device 400 can display a set of target questions to operator 100B and obtain the information filled in by operator 100B regarding the set of target questions. Based on the information filled in, identification device 400 can obtain the risk identification result of the target transaction through the second major model.

[0049] In some embodiments, a first major model can be deployed on server 300, and a second major model can be deployed on identification device 400. When operator 100B triggers a risk identification request for a target transaction on identification device 400, identification device 400 can request a target question set from server 300. Server 300 obtains the transaction information of the target transaction and determines the target question set based on the transaction information of the target transaction using the first major model, and then sends the target question set to identification device 400. The subsequent process of identification device 400 performing risk identification on the target transaction based on the target question set can be referred to the above description and will not be repeated here.

[0050] In some embodiments, both the first and second major models can be deployed on server 300. In this case, when operator 100B triggers a risk identification request for a target transaction on identification device 400, identification device 400 can request a target question set from server 300. Server 300 obtains the transaction information of the target transaction and determines the target question set based on the transaction information using the first major model, then sends the target question set to identification device 400. Identification device 400 displays the target question set to operator 100B and obtains the operator 100B's completed information regarding the target question set. Identification device 400 sends the completed information to server 300 so that server 300 can analyze the completed information using the second major model to obtain the risk identification result of the target transaction. Server 300 can return the risk identification result of the target transaction to identification device 400.

[0051] The identification device 400 is an electronic device with certain data processing capabilities. The identification device 400 can execute the data and instructions of the transaction risk identification method described in this specification, and can execute or be used to execute said data and instructions. For example, the identification device 400 may include hardware devices with data information processing functions and the necessary programs to drive the hardware devices. In some embodiments, the identification device 400 and the server 300 may correspond to the same hardware entity, or the identification device 400 may be a unit / module within the server 300.

[0052] After the identification device 400 obtains the risk identification result of the target transaction, it can send the risk identification result back to the server 300, so that the server 300 can perform subsequent processing on the target transaction based on the risk identification result. For example, if the server 300 determines that the target transaction is a risky transaction based on the risk identification result, it will display a prompt message to the user through the terminal device 200. For example, the prompt message may be "The current transaction has been identified as a risky transaction, and the payment has failed." If the server 300 determines that the target transaction is not a risky transaction based on the risk identification result, it will display a prompt message to the user through the terminal device 200. For example, the prompt message may be "The current transaction is not a risky transaction, and you can continue to the next payment operation."

[0053] It should be understood that Figure 1 The number of terminal devices 200, servers 300, identification devices 400, and training devices 500 shown in the diagram is merely illustrative. Depending on implementation needs, any number of terminal devices 200, servers 300, identification devices 400, and training devices 500 can be included.

[0054] It should be noted that the risk identification method for transactions provided in this manual can be executed entirely on server 300; it can also be executed entirely on identification device 400; or it can be executed partially on server 300 and partially on identification device 400.

[0055] Figure 2 A hardware structure diagram of an electronic device 600 provided according to an embodiment of this specification is shown. The electronic device 600 can be used as... Figure 1 Server 300 in the middle can also be used as Figure 1 The identification device 400 in the middle.

[0056] like Figure 2 As shown, the electronic device 600 may include at least one storage medium 630 and at least one processor 620. In some embodiments, the electronic device 600 may also include a communication port 650 and an internal communication bus 610. Furthermore, the electronic device 600 may also include I / O components 660.

[0057] The internal communication bus 610 can connect to different system components. For example, the internal communication bus 610 can connect to storage medium 630, processor 620, communication port 650, and I / O component 660.

[0058] I / O component 660 supports input / output between electronic device 600 and other components.

[0059] Communication port 650 is used for data communication between electronic device 600 and the outside world. For example, communication port 650 can be used for data communication between electronic device 600 and a network. Communication port 650 can be a wired communication port or a wireless communication port.

[0060] In some embodiments, the network can be any type of wired or wireless network, or a combination thereof. For example, the network may include a cable network, wired network, fiber optic network, telecommunications network, intranet, Internet, local area network (LAN), wide area network (WAN), wireless local area network (WLAN), metropolitan area network (MAN), public switched telephone network (PSTN), Bluetooth network, ZigBee network, near field communication (NFC) network, or similar networks. In some embodiments, the network may include one or more network access points. For example, the network may include wired or wireless network access points, such as base stations or Internet switching points.

[0061] Storage medium 630 may include a data storage device. The data storage device may be a non-transitory storage medium or a temporary storage medium. For example, the data storage device may include one or more of a disk 632, a read-only storage medium (ROM) 634, or a random access storage medium (RAM) 636. Storage medium 630 also includes at least one instruction set stored in the data storage device. The instruction set may include computer program code, which may include programs, routines, objects, components, data structures, processes, modules, etc., that execute the risk identification method for transactions provided in this specification.

[0062] Processor 620 can be communicatively connected to storage medium 630. Processor 620 is used to execute at least one of the above-described instruction sets. When electronic device 600 is operating as server 300 or identification device 400, processor 620 reads the at least one instruction set and executes the transaction risk identification method provided in this specification according to the instructions of the at least one instruction set.

[0063] Processor 620 may be in the form of one or more processors. In some embodiments, processor 620 may include one or more hardware processors, such as microcontrollers, microprocessors, reduced instruction set computers (RISC), application-specific integrated circuits (ASICs), application-specific instruction set processors (ASIPs), central processing units (CPUs), graphics processing units (GPUs), physical processing units (PPUs), microcontroller units, digital signal processors (DSPs), field-programmable gate arrays (FPGAs), advanced RISC machines (ARMs), programmable logic devices (PLDs), any circuit or processor capable of performing one or more functions, etc., or any combination thereof.

[0064] For illustrative purposes, only one processor 620 is shown in the accompanying drawings of electronic device 600. However, it should be noted that electronic device 600 may also include multiple processors. Therefore, the operation and / or method steps disclosed in this specification may be executed by one processor or by multiple processors in combination. For example, if processor 620 of electronic device 600 is described in this specification as performing steps A and B, it should be understood that steps A and B may also be executed jointly or separately by two different processors 620 (e.g., a first processor performs step A, a second processor performs step B, or the first and second processors jointly perform steps A and B).

[0065] Figure 3A flowchart of a transaction risk identification method P100 provided according to an embodiment of this specification is shown. As previously described, the electronic device 600 can execute the transaction risk identification method P100. Specifically, the processor 620 in the electronic device 600 can read an instruction set stored in its local storage medium and then execute the transaction risk identification method P100 according to the provisions of the instruction set.

[0066] like Figure 3 As shown, the risk identification method P100 for transactions may include the following steps.

[0067] S110: In response to a risk identification request for the target transaction, obtain transaction information for the target transaction.

[0068] The target transaction is a transaction awaiting identification. After a user initiates a payment request for the target transaction, the payment platform generates a risk identification task corresponding to the target transaction, and the target transaction enters a state awaiting risk identification. When the risk identification task is assigned to an operator, if the operator triggers risk identification of the target transaction, the electronic device 600 can receive a risk identification request for the target transaction. Alternatively, the electronic device 600 can automatically trigger a risk identification request for the target transaction in the awaiting identification state. In response to the risk identification request, the electronic device 600 obtains the transaction information of the target transaction.

[0069] The transaction information of the target transaction can refer to information related to the target transaction. In some embodiments, the transaction information of the target transaction includes: payment information of the target transaction, proof information of the target transaction, and transaction mode of the target transaction.

[0070] For example, electronic device 600 obtains a payment request corresponding to the target transaction, and then extracts the payment information from the payment request. The payment information includes, but is not limited to, payment time, payment amount, payer, and payee.

[0071] For example, electronic device 600 obtains supporting documents associated with the target transaction and extracts supporting information for the target transaction from these documents. The supporting documents may be uploaded by the user, and the timing of the upload is as described above. If the supporting documents are in image format, electronic device 600 can extract supporting information using Optical Character Recognition (OCR) technology. If the supporting documents are in text format, electronic device 600 can extract supporting information using Named Entity Recognition (NER) technology. Supporting information includes, but is not limited to: the name of the traded item / service, pick-up address, shipping address, logistics information, sender, and recipient.

[0072] For example, electronic device 600 can determine the transaction pattern of the target transaction based on payment information and / or supporting information. Specifically, electronic device 600 can determine the transaction pattern of the target transaction based on the values ​​of one or more fields in the payment information and / or supporting information. Specifically, if the field values ​​in the payment information and / or supporting information match some fields in the definition or description of the target transaction pattern, then electronic device 600 can use the target transaction pattern as the transaction pattern of the target transaction.

[0073] It should be noted that the electronic device 600 can also determine the transaction pattern of the target transaction in other ways. For example, the operator can determine the transaction pattern of the target transaction based on proof information or payment information and input the transaction pattern of the target transaction into the electronic device 600. Alternatively, the user can input the transaction pattern of the target transaction when initiating a payment operation for the target transaction. Other methods may also exist to determine the transaction pattern of the target transaction, which will not be listed here. In this specification, the transaction pattern is used to characterize the type of transaction, including but not limited to general trade patterns, EXW, etc.

[0074] S120: Determine a target problem set based on the transaction information of the target transaction using the first large model. The target problem set includes at least one guiding question for risk identification of the target transaction. The first large model is trained to determine different problem sets for different transactions, and the determined problem set is the minimum problem set that can achieve the risk identification target.

[0075] After obtaining the transaction information of the target transaction, the electronic device 600 can call the first major model and input the transaction information of the target transaction into the first major model, thereby obtaining the target problem set output by the first major model.

[0076] Before calling the primary model, the electronic device 600 has already completed training the primary model. During training, the primary model aims to output an accurate set of questions. An accurate set of questions can be understood as: the number of guiding questions included in the set is the minimum number required to achieve the risk identification objective; or the number of guiding questions included in the set is minimized while still comprehensively covering the key identification points of the current transaction. Achieving the risk identification objective includes, for example, achieving a standard in transaction identification accuracy or a standard in coverage of the key identification points of the transaction.

[0077] The primary model can be trained on electronic devices (e.g., 600 devices) or other devices. There are multiple training methods for the primary model; two of them are described below as examples.

[0078] Training Method 1: The first major model is a selection model, that is, the first major model is trained to select the set of questions corresponding to the transaction to be identified from the question library.

[0079] The first major model can be used Figure 4 P200 is obtained by training as shown. For example, P200 can be executed by training device 500, including the following steps: S210, generate multiple first training samples, wherein each first training sample includes transaction information of historical transactions and a set of historical questions corresponding to the historical transactions, and the set of historical questions includes at least one historical guidance question for risk identification of the historical transactions.

[0080] The historical transactions may be those for which risk identification has been performed by experienced operators. The set of historical questions corresponding to the historical transactions is the set of guiding questions designed by the operators when identifying risks in those historical transactions. For distinction, the guiding questions in the set of historical questions will be referred to as historical guiding questions in the following description. Those skilled in the art will understand that historical guiding questions can reflect / characterize the identification points based on the operators in the process of identifying risks in historical transactions; in other words, historical guiding questions are questions designed by the operators based on the identification points based on the risk identification process.

[0081] It is easy to understand that the operator is an experienced operator, and the set of historical questions designed by this operator is the minimum set of questions that can achieve the risk identification objective. The training device 500 can generate a first training sample based on the transaction information and the set of historical questions mentioned above. Furthermore, the training device 500 can train a first large model based on the first training sample, enabling the first large model to output different sets of questions for different transactions, and the output of the first large model is the minimum set of questions that can achieve the risk identification objective.

[0082] S220, the first large model to be trained was obtained.

[0083] In some embodiments, the training device 500 injects a question library into a first large model to be trained. The question library includes multiple guiding questions, each of which can be derived from the identification points in the SOP. There may be a correspondence between the guiding questions and transaction patterns; for example, each guiding question may correspond to one or more transaction patterns. The first large model to be trained is, for example, a large language model (LLM), which can be obtained based on a base model and serves as the first large model before training. Base models include chatGPT, Llama, etc. Injecting the question library into the first large model to be trained can also be understood as injecting the question library into the first large model.

[0084] It is understood that the identification points of SOP can be converted into the question library by experienced operators, or by other Natural Language Processing (NLP) models or other large models. This specification does not impose any restrictions on this.

[0085] In addition, there are several ways to inject the question library into the first large model to be trained. Two of them are illustrated below.

[0086] Injection Method 1 For example, the training device 500 can obtain the question library locally or on other devices, and generate injection prompt information based on the question library. The injection prompt information includes the question library, and may also include descriptive information. The descriptive information may be, for example, "As a risk control expert, the following is a question library for risk identification. When performing risk identification on a transaction to be identified, you can select a set of questions suitable for the transaction from the question library, as follows..." The injection prompt information is input as a prompt into the first large model to be trained, completing the injection of the question library.

[0087] Injection Method 2 For example, the training device 500 obtains the question library, determines the question sub-libraries corresponding to different transaction modes based on the correspondence between the guiding questions and transaction modes, and generates multiple injection prompt messages. Alternatively, the training device 500 obtains the question sub-libraries corresponding to different transaction modes from other devices and generates the multiple injection prompt messages. Each injection prompt message includes at least one question sub-library corresponding to a transaction mode, and each injection prompt message may also include descriptive information. The descriptive information is, for example, "As a risk control expert, the following is a question sub-library for risk identification of transaction mode #1. When identifying the risk of the transaction to be identified corresponding to transaction mode #1, you can select a set of questions suitable for the transaction to be identified from the question sub-library. The question sub-library is as follows..." Each injection prompt message is input as a prompt into the first large model to be trained to complete the injection of multiple question sub-libraries.

[0088] It is understandable that injection method 2 can overcome the word limit of the prompt, avoiding incomplete injection of the question library due to the word limit, which would affect the first major model's recommendation of the question set.

[0089] S230, with minimizing the first objective function as the training objective, the first large model is trained based on multiple first training samples to obtain the first large model to be trained.

[0090] The first objective function includes a loss term, which represents the difference between the set of prediction questions determined by the first large model based on transaction information from historical transactions and the set of historical questions. The set of prediction questions is selected from the question library by the first large model to be trained based on the transaction information from historical transactions. For example, the training device 500 inputs the transaction information from historical transactions into the first large model to be trained to trigger the first large model to select a portion of guide questions from the question library to form the set of prediction questions.

[0091] For example, any loss function in Natural Language Processing (NLP) can be used as the loss term of the first objective function. For instance, the formula for the first objective function is as follows:

[0092] Where L represents the first objective function, and N1 represents the number of the first training samples in one iteration of training. This represents the set of historical questions in the first training sample i. This represents the set of prediction questions output by the first large model to be trained.

[0093] In some embodiments, the first training sample further includes the historical identification time corresponding to historical transactions. The historical identification time represents the time taken by the operator to perform risk identification on that historical transaction. For example, the time from when the identification device receives a risk identification request to when it obtains a risk identification result can be used as the historical identification time. The historical identification time can also be obtained in other ways, and this specification does not limit this.

[0094] In this case, the first objective function also includes a penalty term associated with the historical identification duration.

[0095] For example, the penalty term can be any type of penalty term that includes a penalty for recognition time. For instance, the formula for the first objective function is as follows:

[0096] in, For the aforementioned penalty item, Indicates the penalty factor. This represents the historical recognition duration in the first training sample i.

[0097] For example, the plurality of first training samples are divided into a training set and a validation set, and the first training samples in the training set are divided into multiple epochs. The training device 500 uses one or more epochs in each round of iterative training, and repeats this process for multiple rounds of iterative training until the training device 500 determines the first largest model obtained from the training based on the validation set and the first objective function, thus reaching the training termination condition.

[0098] It is understandable that if the first training sample does not include historical recognition duration, then the first objective function may not include a penalty term. If the first training sample includes historical recognition duration, then the first objective function may include a penalty term. The guiding questions in the set of prediction questions determined by the first large model obtained through training can be ordered.

[0099] The above solution improves user experience by adding an optimization objective for recognition time to the first objective function, which allows the problem set output by the first large model to correspond to a shorter risk recognition time.

[0100] Training Method 2: The first major model is a generative model, that is, the first major model is trained to generate a set of questions corresponding to the transaction to be identified based on the transaction information of the transaction to be identified.

[0101] The difference between training method 2 and training method 1 is that training method 2 does not require injecting the question library into the first large model to be trained, and the set of prediction questions involved in the first objective function is generated by the first large model based on the transaction information of the historical transactions. For example, during training, after the training device 500 inputs the transaction information of the historical transactions into the first large model to be trained, the first large model to be trained can generate a set of prediction questions based on the transaction information of the historical transactions. The first large model in training method 2 is trained to learn the following ability in supervised learning: to generate (rather than select from the question library) a set of questions that can identify the risk of the transaction based on the transaction information of the transaction to be identified. The remaining training process can be referred to the description in training method 1, and will not be repeated here.

[0102] After the first large model is trained, it can be deployed to electronic device 600.

[0103] For example, the first large model is trained based on training method 1. After obtaining the transaction information of the target transaction, the electronic device 600 generates a first prompt message based on the transaction information of the target transaction and inputs the first prompt message into the first large model. The first prompt message is used to instruct the first large model to select the target question set from the question library based on the transaction information of the target transaction.

[0104] For example, the first prompt information may include transaction information of the target transaction, or the first prompt information may include transaction information of the target transaction and descriptive information. The descriptive information may be, for example, "You are a risk control expert. Please select the minimum set of questions from the question bank that can achieve the risk identification objective based on the following transaction information."

[0105] For example, the first large model is trained based on training method 2. After obtaining the transaction information of the target transaction, the electronic device 600 generates a second prompt message based on the transaction information of the target transaction and inputs the second prompt message into the first large model. The second prompt message is used to instruct the first large model to generate the target question set based on the transaction information of the target transaction.

[0106] For example, the second prompt information includes the transaction information of the target transaction, or the second prompt information includes the transaction information of the target transaction and descriptive information. The descriptive information could be something like, "You are a risk control expert. Please generate a minimum set of questions that can achieve the risk identification objective based on the following transaction information."

[0107] S130: Identify risks in the target transaction based on the target problem set.

[0108] It is understandable that electronic devices can identify risks in target transactions in multiple ways. Three examples are given below.

[0109] Identification Method 1 like Figure 5 As shown, after the first major model outputs the target question set, the electronic device 600 can obtain a preset rule base before performing risk identification on the target transaction. The preset rule base includes preset rules corresponding to each guiding question in the question base. For example, the preset rule base can be stored locally on the electronic device 600; when the electronic device 600 determines the target question set, it can obtain the preset rule base from local storage. Alternatively, the preset rule base can be stored on the server 300; when the electronic device 600 determines the target question set, it can request the preset rule base from the server 300.

[0110] It is understood that the preset rule base may be obtained manually by experienced operators from the aforementioned SOPs, or it may be obtained in other ways. This specification does not limit this.

[0111] The preset rule base includes M preset rules, and each preset rule corresponds to a guiding question, where M is an integer greater than 1. The electronic device 600 can determine multiple preset rules corresponding to all guiding questions in the target question set based on these correspondences. Subsequently, the electronic device 600 can determine the risk identification result of the target transaction based on these multiple preset rules.

[0112] In some embodiments, the guiding questions in the target question set are unordered. The electronic device 600 determines the identification result corresponding to each guiding question in the target question set based on the transaction information of the target transaction and the plurality of preset rules, and determines the risk identification result of the target transaction based on the identification result corresponding to each guiding question.

[0113] For example, the target question set includes guiding questions ①, ②, ③, ④, and ⑤. Among them, Leading question 1: Were the supporting documents altered? Guiding Question 2: Has the seller associated with the target transaction been established for more than X days? Establishment time is, for example, the duration between the seller's business registration date and the time the target transaction was initiated.

[0114] Guiding Question 3: Are the recipient in the proof information and the payer in the payment information the same person or the same legal entity? Guiding Question 4: Are the shipper in the proof information and the payee in the payment information the same person or the same legal entity? Guiding Question 5: Does the seller's number of transactions within day Y exceed the transaction threshold? The multiple preset rules include preset rules corresponding to each guiding question, as follows: The pre-defined rule for guiding question ① is: if the supporting materials are tampered with, the current transaction is a risky transaction; or, if the supporting materials are not tampered with, the current transaction is risk-free in the dimension of guiding question ①.

[0115] The preset rule for the guiding question ② is: if the seller's establishment time is less than or equal to X days, then the current transaction is risky in the dimension of guiding question ②; or if the seller's establishment time is greater than X days, then the current transaction is risk-free in the dimension of guiding question ②.

[0116] The preset rule corresponding to the guiding question ③ is: if the consignee and the payer are the same person or the same legal entity, then the current transaction is risk-free in the dimension of guiding question ③; or, if the consignee and the payer are not the same person or the same legal entity, then the current transaction is risky in the dimension of guiding question ③.

[0117] The preset rule corresponding to the guiding question ④ is: if the shipper and the payee are the same person or the same legal entity, then the current transaction is risk-free in the dimension of guiding question ④; or, if the shipper and the payee are not the same person or the same legal entity, then the current transaction is risky in the dimension of guiding question ④.

[0118] The preset rule for guidance question ⑤ is: if the number of transactions by the seller within Y days is greater than or equal to the transaction number threshold, then the current transaction is at risk in the dimension of guidance question ⑤; or, if the number of transactions by the seller within Y days is less than the transaction number threshold, then the current transaction is not at risk in the dimension of guidance question ⑤.

[0119] In addition, the multiple preset rules may also include: if any of the identification results corresponding to the guiding questions ①, ③ and ④ is risky, then the current transaction is a risky transaction; or if the identification results corresponding to the guiding questions ② and ⑤ are both risky, then the current transaction is a risky transaction.

[0120] Electronic device 600 determines the identification results corresponding to each of the guiding questions ①, ②, ③, ④, and ⑤ based on the transaction information of the target transaction. If any of the identification results corresponding to guiding questions ①, ③, and ④ is risky, then the risk identification result of the target transaction can be determined as risky; or, if the identification results corresponding to guiding questions ② and ⑤ are both risky, then the risk identification result of the target transaction can be determined as risky. If electronic device 600 determines that the identification results corresponding to guiding questions ①, ②, ③, ④, and ⑤ are all risk-free, then the risk identification result of the target transaction can be determined as risky. Alternatively, if electronic device 600 determines that the identification results corresponding to guiding questions ①, ③, and ④ are all risk-free, and at least one of the identification results corresponding to guiding questions ② and ⑤ is risk-free, then the risk identification result of the target transaction can be determined as risky, i.e., the target transaction is a risky transaction.

[0121] In some embodiments, the guiding questions in the target question set are ordered. The electronic device 600 determines the identification result corresponding to each guiding question in sequence according to the order of the guiding questions, and determines the risk identification result of the target transaction based on the identification result corresponding to each guiding question.

[0122] For example, such as Figure 6As shown, the guiding questions in the target question set are, in order, guiding questions ①, ③, ④, ②, and ⑤. Electronic device 600 determines the identification result corresponding to guiding question ① based on the supporting materials of the target transaction and the preset rules corresponding to guiding question ①. If the identification result corresponding to guiding question ① is risky, the target transaction can be determined to be a risky transaction. If the identification result corresponding to guiding question ① is risk-free, then electronic device 600 determines the identification result corresponding to guiding question ③ based on the transaction information of the target transaction and the preset rules corresponding to guiding question ③. If the identification result corresponding to guiding question ③ is risky, the target transaction can be determined to be a risky transaction. If the identification result corresponding to guiding question ③ is risk-free, then electronic device 600 determines the identification result corresponding to guiding question ④ based on the transaction information of the target transaction and the preset rules corresponding to guiding question ④. If the identification result corresponding to guiding question ④ is risky, the target transaction can be determined to be a risky transaction. If the identification result corresponding to guiding question ④ is risk-free, then electronic device 600 determines the identification result corresponding to guiding question ② based on the transaction information of the target transaction and the preset rules corresponding to guiding question ②. If the identification result corresponding to guiding question ② is risk-free, the target transaction can be determined to be a non-risky transaction. If the identification result corresponding to guidance question ② is risky, then electronic device 600 determines the identification result corresponding to guidance question ⑤ based on the transaction information of the target transaction and the preset rules corresponding to guidance question ⑤. If the identification result corresponding to guidance question ⑤ is risky, then the target transaction is a risky transaction; if the identification result corresponding to guidance question ⑤ is risk-free, then the target transaction is not a risky transaction.

[0123] like Figure 6 As shown, the electronic device 600 can sequentially determine the identification result for each guidance question according to its risk level. It first determines the identification result for the guidance question with the highest risk level. If the identification result is risky, there is no need to identify subsequent guidance questions, reducing the waste of the electronic device 600's computing resources and shortening the risk identification time for the target transaction, thus improving the user experience.

[0124] Identification Method 2 In some embodiments, the guiding questions in the target question set are ordered. The electronic device 600 sequentially displays each guiding question in the target question set to the operator, obtains the operator's answers to each guiding question, and determines the risk identification result of the target transaction based on each guiding question and its corresponding answers.

[0125] For example, electronic device 600 displays all or some of the guidance questions ①, ③, ④, ②, and ⑤ sequentially to the operator. Figure 7As shown, if the operator's response to guidance question ① is risky, the electronic device 600 determines the target transaction as a risky transaction and does not need to display guidance questions ③, ④, ②, and ⑤ to the operator. If the operator's response to guidance question ① is risk-free, the electronic device 600 displays guidance question ③ to the operator and obtains the operator's response to guidance question ③. If the response to guidance question ③ is risky, the electronic device 600 determines the target transaction as a risky transaction and does not need to display guidance questions ④, ②, and ⑤ to the operator. If the operator's response to guidance question ③ is risk-free, the electronic device 600 displays guidance question ④ to the operator and obtains the operator's response to guidance question ④. If the response to guidance question ④ is risky, the electronic device 600 determines the target transaction as a risky transaction and does not need to display guidance questions ② and ⑤ to the operator. If the operator's response to guidance question ④ is risk-free, the electronic device 600 displays guidance question ② to the operator and obtains the operator's response to guidance question ②. If the information provided for guidance question ② is risk-free, the electronic device 600 determines that the target transaction is not a risky transaction and does not need to continue displaying guidance question ⑤ to the operator. If the information provided by the operator for guidance question ② is risky, the electronic device 600 further displays guidance question ⑤ to the operator and obtains the information provided by the operator for guidance question ⑤. If the information provided for guidance question ⑤ is risk-free, the electronic device 600 determines that the target transaction is not a risky transaction; if the information provided for guidance question ⑤ is risky, the electronic device 600 determines that the target transaction is a risky transaction.

[0126] like Figure 7 As shown, the electronic device 600 can sequentially determine the information to be filled in for each guidance question according to the risk level of the guidance question. It first determines the information to be filled in for the guidance question with the highest risk level. If the information is deemed risky, there is no need to confirm the information for subsequent guidance questions, reducing the waste of the electronic device 600's computing resources, shortening the risk identification time for the target transaction, and improving the user experience.

[0127] In some embodiments, the guiding questions in the target question set are unordered. The electronic device 600 displays all the guiding questions in the target question set to the operator and obtains the risk identification results filled in by the operator based on all the guiding questions. If the risk identification result filled in by the operator is risky, the electronic device 600 determines that the target transaction is a risky transaction; or, if the risk identification result filled in by the operator is no risk, the electronic device 600 determines that the target transaction is not a risky transaction.

[0128] In some embodiments, the guiding questions in the target question set are unordered. The electronic device 600 displays each guiding question in the target question set to the operator, obtains the operator's answer to each guiding question, and determines the risk identification result of the target transaction based on each guiding question and its corresponding answer.

[0129] For example, after obtaining the information corresponding to each of the guidance questions ①, ②, ③, ④, and ⑤, if any one of the information corresponding to guidance questions ①, ③, and ④ is risky, then the risk identification result of the target transaction can be determined as risky; or, if the information corresponding to guidance questions ② and ⑤ is risky, then the risk identification result of the target transaction can be determined as risky. If the electronic device 600 determines that the information corresponding to guidance questions ①, ②, ③, ④, and ⑤ is risk-free, then the risk identification result of the target transaction can be determined as risky. Alternatively, if the electronic device 600 determines that the information corresponding to guidance questions ①, ③, and ④ is risk-free, and at least one of the information corresponding to guidance questions ② and ⑤ is risk-free, then the risk identification result of the target transaction can be determined as risky, i.e., the target transaction is a risky transaction.

[0130] It is understood that identification method 2 can also have other implementations similar to the above embodiments, which will not be listed one by one.

[0131] Identification Method 3 In some embodiments, such as Figure 8 As shown, after the first large model outputs the target question set, the electronic device 600 displays each guiding question in the target question set to the operator and obtains the operator's input information for each guiding question. Based on each guiding question and its corresponding input information, the electronic device 600 generates third prompt information and inputs this third prompt information into the second large model to obtain the risk identification result of the target transaction. The third prompt information is used to instruct the second large model to perform risk identification on the target transaction. The second large model has the ability to perform risk identification based on the preset rule base and historical risk identification experience, which is learned during the training process of the second large model.

[0132] The second major model can be used Figure 9 The method shown is P300, which is trained. Method P300 can be executed by training device 500 (or it can be executed by other devices; the following explanation uses training device 500 as an example), and includes the following steps: S310, generate multiple second training samples, wherein each second training sample includes: at least one historical guidance question corresponding to the historical transaction, historical filling information corresponding to each of the at least one historical guidance question, and risk labeling information corresponding to the historical transaction.

[0133] Each historical guidance question corresponds to historical completed information, which refers to the operator's answers to that historical guidance question. The risk labeling information represents the risk identification results obtained by the operator through risk assessment of historical transactions. The risk labeling information may indicate whether the transaction is risky or not. A risk labeling of "risky" indicates that the historical transaction was a risky transaction, while a risk labeling of "no risk" indicates that the historical transaction was not a risky transaction.

[0134] S320, obtained the second large model to be trained.

[0135] In some embodiments, the training device 500 injects the preset rule base into the second large model to be trained. The preset rule base is as described above and will not be repeated here.

[0136] In some embodiments, the training device 500 injects both the question library and the preset rule library into the second large model to be trained. The injection method of the question library is as described above and will not be repeated here. The injection process of the preset rule library is described below. There are multiple injection methods, and two of them are illustrated below.

[0137] Injection Method 1 For example, the training device 500 can obtain the preset rule base locally or on other devices, and generate injection prompt information based on the preset rule base. The injection prompt information includes the preset rule base. The injection prompt information may also include descriptive information. For example, the descriptive information might say, "As a risk control expert, the following is a preset rule base for risk identification. When identifying risks in transactions to be identified, you can refer to this preset rule base. The preset rule base is as follows..." The injection prompt information is input as a prompt into the second large model to be trained, completing the injection of the preset rule base. Injecting the preset rule base into the second large model to be trained by the training device 500 can also be understood as injecting the preset rule base into the second large model.

[0138] Injection Method 2 For example, the training device 500 obtains the preset rule library, determines the preset rule sub-libraries corresponding to different transaction modes, and generates multiple injection prompt messages. Alternatively, the training device 500 obtains the preset rule sub-libraries corresponding to different transaction modes from other devices and generates the multiple injection prompt messages. Each injection prompt message includes at least one preset rule sub-library corresponding to a transaction mode. Each injection prompt message may also include descriptive information. The descriptive information is, for example, "As a risk control expert, the following is a preset rule sub-library for risk identification of transaction mode #1. When identifying risks in the transaction to be identified corresponding to transaction mode #1, you can refer to the preset rule sub-library. The preset rule sub-library is as follows..." Each injection prompt message is input as a prompt into the second large model to be trained, completing the injection of multiple preset rule sub-libraries.

[0139] It is understandable that injection method 2 can overcome the word limit of the prompt, avoiding incomplete injection of the preset rule base due to the word limit, which would affect the identification of risky transactions by the second major model.

[0140] S330, with minimizing the second objective function as the training objective, the second large model is trained based on multiple second training samples to obtain the second large model to be trained.

[0141] The second objective function characterizes the difference between the risk prediction information and the risk labeling information of the historical transaction. The risk prediction information is obtained by the second major model based on the at least one historical guidance question and the corresponding historical completion information for each of the at least one historical guidance question, performing risk prediction on the historical transaction. For example, the training device 500 inputs the at least one historical guidance question and its corresponding historical completion information into the second major model, and the second major model outputs the risk prediction information for the historical transaction.

[0142] For example, any NLP loss function can be used as the second objective function, such as the formula for the second objective function as follows:

[0143] Where L represents the second objective function, and N² represents the number of second training samples in one iteration of training. This represents the difference between the risk prediction information and the risk labeling information of the second training sample i. This represents the training label of the second training sample i, with 1 for positive classes and 0 for negative classes. For example, risky is positive and no risk is negative. Let represent the probability that the second training sample i is predicted as positive, with the base of the logarithm being e.

[0144] For example, the plurality of second training samples are divided into a training set and a validation set, and the second training samples in the training set are divided into multiple epochs. The training device 500 uses one or more epochs in each round of iterative training, and repeats this process for multiple rounds of iterative training until the training device 500 determines the second largest model obtained from the training based on the validation set and the second objective function, thus reaching the training termination condition.

[0145] After obtaining the risk identification result of the target transaction through the identification method described above, if the risk identification result indicates that the target transaction is a risky transaction, the electronic device 600 can send a risk warning message to the terminal device 200. The risk warning message may be, for example, "After identification, the current transaction is a risky transaction, and payment has failed." The terminal device 200 can display the risk warning message to the user associated with the target transaction to inform the user of the payment failure. If the risk identification result indicates that the target transaction is not a risky transaction, the electronic device 600 can send a payment warning message to the terminal device 200. The payment warning message may be, for example, "After identification, the current transaction is not a risky transaction; please continue with the subsequent payment process." The terminal device 200 can display the payment warning message to the user associated with the target transaction to guide the user to the next step of the payment process.

[0146] Method 3 employs a second major model. By injecting a pre-defined rule base into this model and leveraging its language understanding capabilities, it ensures that risks explicitly prohibited by the pre-defined rule base are not overlooked during risk identification. Furthermore, because the second major model can learn risk identification strategies beyond the pre-defined rule base during supervised fine-tuning (SFT) training, its identification of risky transactions is more comprehensive and accurate. Additionally, using the second major model avoids the randomness caused by different operators using varying standards to identify transactions.

[0147] In summary, the transaction risk identification method, device, and storage medium provided in this specification enable the electronic device 600 to determine the target question set for identifying a target transaction before performing risk identification. Subsequently, risk identification is performed sequentially based on guiding questions from this target question set, avoiding omissions of key identification points and improving the accuracy of risk identification. Furthermore, the number of guiding questions included in the target question set is the minimum required to achieve the risk identification objective, preventing the electronic device 600 from identifying redundant guiding questions during the identification process, which would lead to excessively long identification times and negatively impact user experience. Moreover, the guiding questions in the target question set will differ for different transactions, allowing the electronic device 600 to more specifically identify risky transactions, offering greater flexibility. Additionally, using the first major model lowers the learning curve for operators, reducing training costs for payment platforms.

[0148] This specification, in another aspect, provides a computer-readable non-transitory storage medium storing at least one set of executable instructions for risk identification in transactions. When the executable instructions are executed by a processor, they instruct the processor to perform the steps of method P100 described herein. In some possible embodiments, various aspects of this specification can also be implemented as a program product comprising program code. When the program product is run on electronic device 600, the program code causes electronic device 600 to perform the steps of method P100 described herein. The program product for implementing the above method may employ a portable compact disc read-only memory (CD-ROM) containing program code and may run on electronic device 600. However, the program product of this specification is not limited thereto. In this specification, the readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system. The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. The computer-readable storage medium may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium may also be any readable medium other than a readable storage medium that can send, propagate, or transmit programs for use by or in connection with an instruction execution system, apparatus, or device. Program code contained on a readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof. Program code for performing the operations described herein can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on electronic device 600, partially on electronic device 600, as a standalone software package, partially on electronic device 600 and partially on a remote computing device, or entirely on a remote computing device.

[0149] Unless otherwise stated, the term "multiple" in this specification shall be understood as two or more.

[0150] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0151] In summary, after reading this detailed disclosure, those skilled in the art will understand that the foregoing detailed disclosure is presented by way of example only and is not restrictive. Although not explicitly stated herein, those skilled in the art will understand that this specification requires various reasonable changes, improvements, and modifications to the embodiments. These changes, improvements, and modifications are intended to be made by this specification and are within the spirit and scope of the exemplary embodiments described herein.

[0152] Furthermore, certain terms in this specification have been used to describe embodiments of this specification. For example, "an embodiment," "an embodiment," and / or "some embodiments" mean that a particular feature, structure, or characteristic described in connection with that embodiment may be included in at least one embodiment of this specification. Therefore, it is to be emphasized and understood that two or more references to "an embodiment" or "an embodiment" or "alternative embodiment" in various parts of this specification do not necessarily refer to the same embodiment. Moreover, specific features, structures, or characteristics may be suitably combined in one or more embodiments of this specification.

[0153] It should be understood that in the foregoing description of the embodiments in this specification, various features are combined in a single embodiment, drawing, or description for the purpose of simplifying the description and aiding in the understanding of a feature. However, this does not mean that the combination of these features is necessary, and those skilled in the art may readily identify some of the devices as separate embodiments when reading this specification. That is, the embodiments in this specification can also be understood as an integration of multiple secondary embodiments. It is also valid when each secondary embodiment contains fewer than all the features of a single foregoing disclosed embodiment.

[0154] Every patent, patent application, publication of a patent application, and other material such as articles, books, specifications, publications, documents, articles, etc., cited herein, except for those inconsistent with or conflicting with this document, or those having a restrictive effect on the widest scope of the claims, may be incorporated herein by reference for all purposes now or hereafter associated with this document. Furthermore, in the event of any inconsistency or conflict between the description, definition, and / or use of relevant terms in any material and the description, definition, and / or use of relevant terms in this document, the terms in this document shall prevail.

[0155] Finally, it should be understood that the embodiments disclosed herein are illustrative of the principles of the embodiments described in this specification. Other modified embodiments are also within the scope of this specification. Therefore, the embodiments disclosed in this specification are merely examples and not limitations. Those skilled in the art can implement the applications described in this specification using alternative configurations based on the embodiments in this specification. Therefore, the embodiments in this specification are not limited to the embodiments precisely described in the applications.

Claims

1. A method for identifying risks in a transaction, comprising: In response to a risk identification request for a target transaction, obtain the transaction information of the target transaction; The first major model determines a set of target questions based on the transaction information of the target transaction. The set of target questions includes at least one guiding question for risk identification of the target transaction. The first major model is trained to determine different sets of questions for different transactions, and the set of questions is the minimum set of questions that can achieve the risk identification target. as well as Risk identification is performed on the target transaction based on the target problem set.

2. The method according to claim 1, wherein, The transaction information of the target transaction includes the payment information of the target transaction, the transaction mode of the target transaction, and the proof information of the target transaction.

3. The method according to claim 2, wherein, The process of obtaining the transaction information of the target transaction includes: Obtain the payment request corresponding to the target transaction, and extract the payment information of the target transaction from the payment request; Obtain supporting documentation associated with the target transaction, and extract supporting information about the target transaction from the supporting documentation; and Based on the payment information and / or the proof information, the transaction mode of the target transaction is determined.

4. The method according to claim 1, wherein, The first large model is obtained by training multiple first training samples with the goal of minimizing a first objective function. Each first training sample includes transaction information of historical transactions and a set of historical questions corresponding to the historical transactions. The set of historical questions includes at least one historical guiding question for risk identification of the historical transactions. The first objective function includes a loss term, which represents the difference between the set of prediction questions determined by the first large model based on the transaction information of the historical transactions and the set of historical questions.

5. The method according to claim 4, wherein, Each first training sample also includes the historical identification duration corresponding to the historical transaction, and The first objective function also includes a penalty term, which is associated with the historical recognition duration.

6. The method according to claim 4, wherein, The first large model is injected with a question library, and the set of prediction questions is selected by the first large model from the question library based on the transaction information of the historical transactions.

7. The method according to claim 6, wherein, The process of determining the target problem set based on the transaction information of the target transaction using the first major model includes: Generate a first notification message based on the transaction information of the target transaction; and The first prompt information is input into the first large model to instruct the first large model to select the target question set from the question library based on the transaction information of the target transaction.

8. The method according to claim 4, wherein, The first large model is a generative model, and the set of prediction questions is generated by the first large model based on the transaction information of the historical transactions.

9. The method according to claim 8, wherein, The process of determining the target problem set based on the transaction information of the target transaction using the first major model includes: A second notification message is generated based on the transaction information of the target transaction; and The second prompt information is input into the first large model to instruct the first large model to generate the target question set based on the transaction information of the target transaction.

10. The method according to claim 1, wherein, The risk identification of the target transaction based on the target problem set includes: Show the operator the at least one guidance question and obtain the operator's response to the at least one guidance question; and Based on the at least one guiding question and its corresponding information, the risk identification result of the target transaction is determined.

11. The method of claim 10, wherein, The process of determining the risk identification result of the target transaction based on the at least one guiding question and its corresponding filled-in information includes: A third prompt message is generated based on the at least one guiding question and its corresponding filled-in information; and The third prompt information is input into the second large model to instruct the second large model to perform risk identification on the target transaction and obtain the risk identification result of the target transaction. The second large model has the ability to perform risk identification based on a preset rule base and historical risk identification experience, which is learned during the training process of the second large model.

12. The method according to claim 11, wherein, The second major model is trained based on multiple second training samples, with the goal of minimizing a second objective function. Each second training sample includes at least one historical guidance question corresponding to a historical transaction, historical completion information corresponding to each of the at least one historical guidance question, and risk labeling information corresponding to the historical transaction. The second objective function characterizes the difference between the risk prediction information and the risk labeling information of the historical transactions. The risk prediction information is obtained by the second major model based on the risk prediction of the historical transactions using the at least one historical guidance question and the historical filling information corresponding to each of the at least one historical guidance question.

13. An electronic device, comprising: At least one storage medium storing at least one instruction set for risk identification in transactions; as well as At least one processor is communicatively connected to the at least one storage medium, wherein the at least one processor reads the at least one instruction set during operation and, according to the instructions of the at least one instruction set: In response to a risk identification request for the target transaction, the transaction information of the target transaction is obtained. A first-level model determines a set of target questions based on the transaction information of the target transaction. This set includes at least one guiding question for risk identification of the target transaction. The first-level model is trained to determine different sets of questions for different transactions, and these sets are the minimum set of questions capable of achieving the risk identification objective. Risk identification is performed on the target transaction based on the target problem set.

14. The electronic device according to claim 13, wherein, The transaction information of the target transaction includes the payment information of the target transaction, the transaction mode of the target transaction, and the proof information of the target transaction.

15. The electronic device according to claim 14, wherein, In order to obtain the transaction information of the target transaction, the at least one processor: Obtain the payment request corresponding to the target transaction, and extract the payment information of the target transaction from the payment request; Obtain supporting documentation related to the target transaction, and extract supporting information about the target transaction from the supporting documentation; as well as Based on the payment information and / or the proof information, the transaction mode of the target transaction is determined.

16. The electronic device according to claim 13, wherein, The first large model is obtained by training multiple first training samples with the goal of minimizing a first objective function. Each first training sample includes transaction information of historical transactions and a set of historical questions corresponding to the historical transactions. The set of historical questions includes at least one historical guiding question for risk identification of the historical transactions. The first objective function includes a loss term, which represents the difference between the set of prediction questions determined by the first large model based on the transaction information of the historical transactions and the set of historical questions.

17. The electronic device according to claim 16, wherein, Each first training sample also includes the historical review duration corresponding to the historical transaction, and The first objective function also includes a penalty term, which is associated with the historical review duration.

18. The electronic device according to claim 16, wherein, The first large model is injected with a question library, and the set of prediction questions is selected by the first large model from the question library based on the transaction information of the historical transactions.

19. The electronic device according to claim 18, wherein, To determine the target problem set, the at least one processor: A first notification message is generated based on the transaction information of the target transaction; as well as The first prompt information is input into the first large model to instruct the first large model to select the target question set from the question library based on the transaction information of the target transaction.

20. A computer-readable non-transitory storage medium, wherein, The computer-readable non-transitory storage medium stores at least one set of instructions, which, when executed by at least one processor, implement the method as described in any one of claims 1-12.

Images