Financial service processing method, device, storage medium and program product

Abstract

The application discloses a kind of financial service processing method, device, storage medium and program product.It relates to artificial intelligence and financial technology field, the method comprises: receiving the update demand submitted by first object, wherein the first object includes: the object of financial institution for customer to handle financial business, financial intelligent system includes: M intelligent agent for processing financial business and target language model;The update demand is analyzed, and the analysis result is obtained, and training task is generated based on the analysis result;Based on training task, update financial intelligent system using sandbox model, obtain updated financial intelligent system, wherein sandbox model includes: the copy of the model already deployed in financial intelligent system isolated from the production environment of financial intelligent system.Through the present application, the problem of low efficiency of new update demand landing in related art by developing team to develop update demand in financial business system is solved.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and financial technology, and more specifically, to a financial business processing method, apparatus, storage medium, and program product. Background Technology

[0002] In related technologies, the intelligent systems of financial institutions (such as customer service robots and risk control models) mainly rely on fixed training models or manual rule engines, which have the following technical shortcomings:

[0003] (1) Business and technology are disconnected: Model / rule updates require the intervention of the algorithm / development team, and the implementation cycle of business requirements can take several weeks, making it impossible to respond quickly to changes in requirements;

[0004] (2) High maintenance costs: Model iteration requires retraining the entire dataset, and rule adjustments require manual code rewriting, resulting in high manpower and time costs.

[0005] There is currently no effective solution to the problem of low efficiency in implementing new update requirements in financial business systems through development teams in related technologies. Summary of the Invention

[0006] The main purpose of this application is to provide a financial business processing method, device, storage medium and program product to solve the problem of low efficiency in implementing new update requirements in financial business systems through development teams in related technologies.

[0007] To achieve the above objectives, according to one aspect of this application, a financial transaction processing method is provided. The method includes: receiving an update request submitted by a first object, wherein the first object includes an object in a financial institution that handles financial transactions for customers, the update request including a request to update the business processing functions of a financial intelligent system, the financial intelligent system including M intelligent agents for processing the financial transactions and a target language model, where M is a positive integer; parsing the update request to obtain a parsing result, and generating a training task based on the parsing result; updating the financial intelligent system using a sandbox model based on the training task to obtain an updated financial intelligent system, wherein the sandbox model includes a copy of the already deployed model in the financial intelligent system, isolated from the production environment of the financial intelligent system.

[0008] Further, based on the training task, the financial intelligence system is updated using a sandbox model to obtain an updated financial intelligence system, including: training the sandbox model based on the training task until the sandbox model meets a preset convergence condition to obtain a trained sandbox model; using an A / B testing strategy to test the financial intelligence system and the trained sandbox model to obtain test results; and updating the financial intelligence system based on the trained sandbox model when the test results indicate that the accuracy of the trained sandbox model in processing the financial business is higher than that of the financial intelligence system.

[0009] Furthermore, after updating the financial intelligence system using a sandbox model based on the training task to obtain the updated financial intelligence system, the method further includes: conducting a canary release of the updated financial intelligence system based on a preset ratio; and recording a snapshot of the updated financial intelligence system and the update requirement in the audit log after the canary release of the updated financial intelligence system is completed.

[0010] Furthermore, after updating the financial intelligence system using a sandbox model based on the training task to obtain the updated financial intelligence system, the process further includes: receiving a business processing request submitted by a second object, and performing structured processing on the business processing request to obtain an intermediate representation, wherein the second object includes an object that receives services from the financial institution; determining S sub-tasks based on the intermediate representation, wherein each sub-task is associated with one of the intelligent agents, and S is a positive integer less than M; and processing the business for the second object based on the intelligent agents associated with the S sub-tasks to obtain a business processing result.

[0011] Furthermore, the financial intelligence system includes: a knowledge memory layer, which stores processing rules for processing the financial business, and processing the business for the second object based on S intelligent agents associated with the sub-tasks to obtain business processing results, including: based on the S intelligent agents associated with the sub-tasks, calling the processing rules in the knowledge memory layer to process the intermediate representation to obtain S processing results; prioritizing the S processing results to obtain S sorted processing results; and generating a business processing result based on the S sorted processing results and the target language model.

[0012] Furthermore, after generating the business processing result based on the S sorted processing results, the method further includes: generating reply content based on a preset template when the business processing result indicates that preset information needs to be supplemented; and reviewing the content supplemented by the second object after the second object has supplemented the preset information based on the reply content.

[0013] Furthermore, after processing the business for the second object based on the S intelligent agents associated with the sub-tasks and obtaining the business processing result, the method further includes: obtaining the call records and rule call data of the S intelligent agents associated with the sub-tasks, wherein the rule call data includes: data of the rules called from the knowledge memory layer by the S intelligent agents associated with the sub-tasks; and generating an evidence chain based on the call records, the rule call data, and the intermediate representation.

[0014] To achieve the above objectives, according to another aspect of this application, a financial transaction processing apparatus is provided. The apparatus includes: a receiving unit for receiving an update request submitted by a first object, wherein the first object includes an object in a financial institution that handles financial transactions for customers, and the update request includes a request to update the business processing functions of a financial intelligent system, wherein the financial intelligent system includes M intelligent agents for processing the financial transactions and a target language model, where M is a positive integer; a first processing unit for parsing the update request, obtaining a parsing result, and generating a training task based on the parsing result; and a second processing unit for updating the financial intelligent system using a sandbox model based on the training task, to obtain an updated financial intelligent system, wherein the sandbox model includes a copy of the already deployed model in the financial intelligent system, isolated from the production environment of the financial intelligent system.

[0015] Further, the second processing unit includes: a training subunit, used to train the sandbox model based on the training task until the sandbox model meets a preset convergence condition, thereby obtaining a trained sandbox model; a testing subunit, used to test the financial intelligence system and the trained sandbox model using an A / B testing strategy, and obtain test results; and an updating subunit, used to update the financial intelligence system based on the trained sandbox model when the test results indicate that the accuracy of the trained sandbox model in processing the financial business is higher than that of the financial intelligence system.

[0016] Furthermore, the financial business processing device also includes: a publishing unit, used to update the financial intelligence system using a sandbox model based on the training task, and after obtaining the updated financial intelligence system, to perform a gray-scale release of the updated financial intelligence system based on a preset ratio; and a recording unit, used to record a snapshot of the updated financial system and the update requirement to an audit log when the gray-scale release of the updated financial intelligence system is completed.

[0017] Furthermore, the financial business processing device further includes: a receiving unit, configured to receive a business processing request submitted by a second object after updating the financial intelligent system using a sandbox model based on the training task to obtain an updated financial intelligent system, and to perform structured processing on the business processing request to obtain an intermediate representation, wherein the second object includes an object being provided services by the financial institution; a determining unit, configured to determine S sub-tasks based on the intermediate representation, wherein each sub-task is associated with one of the intelligent agents, and S is a positive integer less than M; and a third processing unit, configured to process the business for the second object based on the intelligent agents associated with the S sub-tasks, and obtain a business processing result.

[0018] Furthermore, the financial intelligence system includes: a knowledge memory layer storing processing rules for processing the financial business; and a third processing unit comprising: a processing subunit for processing the intermediate representation based on S intelligent agents associated with the subtasks, invoking the processing rules in the knowledge memory layer to obtain S processing results; a sorting subunit for prioritizing the S processing results to obtain S sorted processing results; and a generation subunit for generating a business processing result based on the S sorted processing results and the target language model.

[0019] Furthermore, the third processing unit also includes: a generation subunit, used to generate a response content based on a preset template after generating a business processing result based on the S sorted processing results, when the business processing result indicates that preset information needs to be supplemented; and an auditing subunit, used to audit the content supplemented by the second object after the second object has supplemented the preset information based on the response content.

[0020] Furthermore, the financial business processing device further includes: an acquisition unit, configured to acquire, after processing the business for the second object based on the S intelligent agents associated with the sub-tasks and obtaining the business processing result, the acquisition unit acquires the call records and rule call data of the S intelligent agents associated with the sub-tasks, wherein the rule call data includes: data of rules called from the knowledge memory layer by the S intelligent agents associated with the sub-tasks; and a generation unit, configured to generate an evidence chain based on the call records, the rule call data, and the intermediate representation.

[0021] According to another aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is executed, it controls the device where the computer-readable storage medium is located to perform the financial business processing method.

[0022] According to another aspect of this application, an electronic device is provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the program executes the financial transaction processing method when it runs.

[0023] According to another aspect of this application, a computer program product is provided, including computer instructions that, when executed by a processor, implement the steps of the financial business processing method.

[0024] In this application, the following approach is adopted: receiving update requests submitted by a first object, wherein the first object includes an object in a financial institution that handles financial business for customers; the update request includes a request to update the business processing functions of a financial intelligent system; the financial intelligent system includes M intelligent agents for processing financial business and a target language model, where M is a positive integer; parsing the update request to obtain the parsing result, and generating a training task based on the parsing result; and updating the financial intelligent system using a sandbox model based on the training task to obtain an updated financial intelligent system, wherein the sandbox model includes a copy of the already deployed model in the financial intelligent system that is isolated from the production environment of the financial intelligent system. This solves the technical problem in related technologies where update requests are developed and implemented by a development team in the financial business system, resulting in low efficiency in implementing new update requests. In this application, based on the update requests input by the business handling object, the model in the financial intelligent system is adjusted using a sandbox model, avoiding the situation in related technologies where a development team is required to develop and implement update requests, resulting in low efficiency. This achieves the technical effect of improving the efficiency of implementing business requirements of the financial system. Attached Figure Description

[0025] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0026] Figure 1 A hardware structure block diagram of a computer terminal for implementing financial business processing methods is shown.

[0027] Figure 2 This is a flowchart of a financial business processing method provided according to an embodiment of this application;

[0028] Figure 3 This is a schematic diagram of a financial business processing system provided according to an embodiment of this application;

[0029] Figure 4 This is a schematic diagram of the update and optimization process provided according to the embodiments of this application;

[0030] Figure 5This is a timing diagram of intelligent agent collaboration for loan applications provided according to embodiments of this application;

[0031] Figure 6 This is a schematic diagram of a financial transaction processing apparatus provided according to an embodiment of this application;

[0032] Figure 7 This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0033] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0034] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0035] It should be noted that the financial business processing method and apparatus in this application can be used in the financial technology field to directly update and correct models / rules based on the needs of business personnel, and can also be used in any field other than the financial technology field to directly update and correct models / rules based on the needs of business personnel. This application does not limit the application field of the financial business processing method and apparatus.

[0036] First, some nouns or terms that appear in the description of the embodiments of this application shall be interpreted as follows:

[0037] Target Intelligence System: An AI system that allows business users to directly optimize models / rules through examples, error correction, and rule input, without relying on the algorithm team to make code-level adjustments.

[0038] Intermediate Representation (IR): A structured data format for unified multimodal input (text, voice, form), containing core fields such as business intent, entity information, and context, providing standardized data interfaces for each module.

[0039] Update requirements: A structured storage unit for example / correction / rule information input by business personnel, including metadata such as update type, input / output, applicable scenarios, and compliance tags.

[0040] Sandbox model: A copy of the model isolated from the production environment, used for offline fine-tuning based on update requirements, avoiding direct impact on actual business operations.

[0041] Decision Evidence Chain: Records the entire chain of information for business requests from input to output (including input IR, agent invocation, rule triggering, and model version) for auditing and interpretability traceability.

[0042] It should be noted that the information collected in this application (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of this data all comply with relevant laws and standards, have taken necessary security measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse. For example, this system has interfaces with relevant users or organizations, providing users with corresponding operation entry points for users to choose to agree to or refuse automated decision results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.

[0043] This invention can be applied to various software products, control systems, and client-side (including but not limited to: mobile clients, PCs, etc.) control systems of financial institutions. It can update and fine-tune the models deployed in the system, and accurately process the business content of financial institutions (including but not limited to: transfer, wealth management, funds, payment, account inquiry, advertising, recommendation and other business functions).

[0044] Example 1

[0045] According to an embodiment of this application, a method embodiment for financial business processing is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0046] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1A hardware structure block diagram of a computer terminal (or mobile device) for implementing financial business processing methods is shown. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0047] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0048] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the financial business processing method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned financial business processing method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0049] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0050] The display can be, for example, a touchscreen liquid crystal display (LCD), which allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0051] Under the aforementioned operating environment, this application provides the following: Figure 2 The financial transaction processing method shown. Figure 2 This is a flowchart of a financial business processing method according to Embodiment 1 of this application.

[0052] Step S201: Receive the update request submitted by the first object, wherein the first object includes: an object in a financial institution that handles financial business for customers, and the update request includes: a request to update the business processing function of the financial intelligent system, wherein the financial intelligent system includes: M intelligent agents for processing financial business and a target language model, where M is a positive integer.

[0053] The first target mentioned above can refer to non-technical personnel (such as business operations personnel) with business permissions within financial institutions. The update requirements mentioned above can refer to instructions targeting the optimization of existing financial intelligent system functions, such as examples of financial business processing, error correction, and processing rules, to improve the system's responsiveness to the latest business processing needs. M intelligent agents can work in parallel within the system, responsible for executing different business processing tasks. For example, the M intelligent agents can include: a customer service intelligent agent for authentication, a risk control intelligent agent for risk assessment, and a compliance intelligent agent for compliance detection.

[0054] The target language model can refer to a large language model or a special instruction model that supports the above-mentioned intelligent agent in semantic understanding, reasoning and decision-making. It is the underlying reasoning engine when the intelligent agent performs tasks.

[0055] In one alternative example, an update request submitted by a first object through a visual interface can be received, which may indicate the functionalities that need to be optimized.

[0056] Step S202: Analyze the update requirements, obtain the analysis results, and generate training tasks based on the analysis results.

[0057] The above parsing results refer to the structured data obtained by structuring unstructured update requirements (e.g., natural language or sample data), such as the "event-triggering condition-action" structure.

[0058] The “training task” mentioned above refers to the job package used to drive the iteration of the model or rule engine, which may include input and output samples, constraints and expected goals.

[0059] In one optional example, for received update requests, natural language processing techniques can be used to perform deep parsing of their content. This involves identifying business objectives through intent recognition, extracting key entities and logical relationships, and transforming unstructured update requests into structured data, forming a parsing result. Subsequently, based on this parsing result and pre-defined mapping rules, corresponding training tasks can be automatically generated. For example, the training type (such as fine-tuning or rule injection) can be determined, and a training dataset can be constructed. The final generated training task can trigger subsequent offline fine-tuning in a sandbox environment, ensuring that the intuitive needs of business personnel can be accurately translated into executable instructions for the model.

[0060] Step S203: Based on the training task, the financial intelligence system is updated using a sandbox model to obtain the updated financial intelligence system. The sandbox model includes a copy of the already deployed model in the financial intelligence system that is isolated from the production environment of the financial intelligence system.

[0061] A sandbox model can refer to a copy of a model that is physically or logically isolated from the production environment. It inherits the weights and configurations of the currently effective version in the production environment and is used for offline fine-tuning in a secure, isolated manner. An updated financial intelligence system refers to a financial intelligence system that has undergone sandbox verification, passed compliance checks, and has been switched to the new model version, possessing optimized business processing capabilities.

[0062] In one optional example, a snapshot of the current active model in the production environment of the financial intelligence system can be obtained. A sandbox model is then instantiated in the sandbox environment to ensure the training baseline is consistent with the production environment. Subsequently, the training task can be deployed to the sandbox to drive offline fine-tuning of the target language model or rule engine. During this process, the sandbox model independently maintains its training state, blocking connections with production data and traffic gateways to ensure production environment stability. After fine-tuning, the convergence of the loss function and compliance metrics are checked. If the checks pass, the model in the financial intelligence system can be updated based on the trained sandbox model, resulting in an updated financial intelligence system that enables rapid implementation of business requirements. If the checks fail, the update process can be terminated.

[0063] Through the above steps, in this embodiment, based on the update requirements input by the business processing object, the model in the financial intelligent system is adjusted using a sandbox model. This avoids the inefficient situation in related technologies where the development team needs to develop and implement update requirements on their own, thus improving the efficiency of developing and implementing business requirements for the financial system. Furthermore, it solves the technical problem in related technologies where the development team has to develop and implement update requirements in the financial business system, resulting in low efficiency in implementing new update requirements.

[0064] An optional financial transaction processing system can be used to execute the financial transaction processing method provided in this embodiment. Figure 3 This is a schematic diagram of a financial business processing system provided according to an embodiment of this application, such as... Figure 3 As shown, it includes:

[0065] (1). Multimodal access layer: receives input such as text, voice, and forms, and converts them into a unified intermediate representation (IR);

[0066] (2). Update management layer: Process the examples / corrections / rules input by business personnel, generate update requirements and trigger model / rule updates;

[0067] (3). Knowledge and memory layer: structured storage of update requirements, knowledge graphs, and business rules; all data is labeled with version and source tags.

[0068] (4). Intelligent Agent Collaboration Layer: Manages intelligent agents in areas such as customer service and risk control, breaks down tasks, invokes capabilities, resolves conflicts, and aggregates results;

[0069] (5). Reasoning and Model Layer: Includes large language model (corresponding to target language model), domain instruction model, rule engine, and performs reasoning based on IR and knowledge;

[0070] (6) Audit and security layer: Generate decision evidence chain and provide encrypted storage, access control and model rollback capabilities.

[0071] Figure 4 This is a schematic diagram of the update and optimization process provided according to the embodiments of this application, such as... Figure 4 As shown, it includes:

[0072] (1) The first target (e.g., business personnel) submits update requests (examples / corrections / rules) through a visual interface;

[0073] (2). The management layer updates the records and parses them into an "event-triggering condition-execution action" structure to generate training tasks;

[0074] (3). The sandbox model is fine-tuned offline based on the training task. After verifying the convergence of the loss function, A / B testing is started.

[0075] (4). If the KPIs (accuracy rate, compliance trigger rate) of the sandbox model are better than those of the production model in the A / B test, the model will be gradually rolled out in a phased manner at a rate of 10% → 30% → 100%.

[0076] (5) After full deployment, record model snapshots and update requirement relationships to the audit log.

[0077] Optionally, in the financial business processing method provided in this application embodiment, the financial intelligence system is updated using a sandbox model based on a training task to obtain an updated financial intelligence system. This includes: training the sandbox model based on the training task until the sandbox model meets a preset convergence condition to obtain a trained sandbox model; using an A / B testing strategy to test the financial intelligence system and the trained sandbox model to obtain test results; and updating the financial intelligence system based on the trained sandbox model when the test results indicate that the accuracy of the trained sandbox model in processing financial business is higher than that of the financial intelligence system.

[0078] The aforementioned pre-defined convergence conditions can refer to the loss function value decreasing to a threshold or the validation set metrics meeting certain standards during training, ensuring model stability. A / B testing can involve routing a portion of real business traffic to both the original production model and the new sandbox model, and then evaluating the new model's performance by comparing key performance indicators (such as accuracy and compliance trigger rate). The test results can be a quantitative evaluation report based on the A / B testing strategy, used to determine whether the new model is superior to the old model.

[0079] In one optional example, the training task can be performed offline in a sandbox environment to train the sandbox model. The training process continuously monitors the loss curve and validation set performance until a preset convergence condition is met, generating a post-trained sandbox model with optimization capabilities. Then, A / B testing can be initiated, proportionally (e.g., 10%) diverting real business traffic to the production environment and the sandbox model. Data on the response accuracy, latency, and compliance of both models when processing financial transactions is collected, and comparative test results are generated. If the test results clearly indicate that the post-trained sandbox model significantly outperforms the production environment model in terms of accuracy and other metrics, the update is deemed effective. Finally, a canary release process is triggered, gradually increasing the proportion of traffic to the sandbox model (e.g., sequentially increasing to 30%, then 100%) until a full switch is completed, thus updating the financial intelligence system, ensuring the security and effectiveness of business function iteration, and achieving a seamless transition from sandbox validation to production deployment.

[0080] Optionally, in the financial business processing method provided in the embodiments of this application, after updating the financial intelligence system based on the training task and using the sandbox model to obtain the updated financial intelligence system, the method further includes: performing a gray-scale release of the updated financial intelligence system based on a preset ratio; and recording the snapshot of the updated financial system and the update requirements to the audit log after the gray-scale release of the updated financial intelligence system is completed.

[0081] Gray-scale release refers to gradually releasing an updated model to users at a predetermined rate (e.g., 10%, 30%, 100%) to verify the stability of the new model with low risk. Snapshots of the updated financial system can be used for rapid rollback. Audit logs are immutable logs that record end-to-end decision-making information, including update requests, model versions, and execution results, meeting the traceability requirements of financial regulations.

[0082] In one optional example, the updated financial intelligence system can be released in a canary rollout, gradually increasing the traffic share of the updated system according to a preset ratio, while monitoring for anomalies in business metrics in real time. The canary rollout is complete once the traffic has been fully switched over and the system is running stably. For the updated financial intelligence system, a snapshot of the current model state can also be generated, fixing version information and key parameters. The original update request, associated parsing results, training tasks, model version, and release process logs can also be written to the audit log. This not only ensures the traceability of model changes, meeting the financial regulatory requirements of "explainable decisions and auditable end-to-end processes," but also guarantees rapid rollback capabilities in extreme situations through the snapshot mechanism, achieving a secure closed loop for business iteration.

[0083] Optionally, in the financial business processing method provided in this application embodiment, after updating the financial intelligence system based on the training task using a sandbox model to obtain the updated financial intelligence system, the method further includes: receiving a business processing request submitted by a second object, and performing structured processing on the business processing request to obtain an intermediate representation, wherein the second object includes: an object that receives services from a financial institution; determining S sub-tasks based on the intermediate representation, wherein each sub-task is associated with one intelligent agent, and S is a positive integer less than M; and processing the business for the second object based on the intelligent agents associated with the S sub-tasks to obtain the business processing result.

[0084] The second object can refer to the recipient of financial services, such as a customer. A business processing request can be a specific financial need initiated by the second object, such as a loan application or consultation. The S subtasks can be the smallest units of work that need to be executed by a specific intelligent agent, decomposed from the intermediate representation (IR). The business processing result can be the processing results fed back by each intelligent agent.

[0085] In one optional example, a business processing request submitted by a second object can be received, cleaned and standardized by a multimodal access layer, and transformed into an intermediate representation containing core business fields. Then, the agent collaboration layer parses the IR (Intent Relationship) and, based on the intent matching capability routing matrix, decomposes it into S structured subtasks, each bound to a specific target agent. After being orchestrated according to dependencies, the subtasks can be distributed to the corresponding agents in parallel or sequentially. Each agent loads the latest model or rules, executes inference, and returns results and evidence chain fragments. The collaboration layer can perform conflict detection and aggregation on the results; if contradictions exist, it can trigger manual review based on priority strategies. Finally, a matching decision template generates a business processing result, which is returned to the second object, and the entire process is recorded in the audit log to ensure efficient and traceable response.

[0086] One possible business reasoning process includes:

[0087] (1). The multimodal access layer converts user requests into intermediate representations (IR);

[0088] (2). The agent collaboration layer splits the IR into sub-tasks (such as intent recognition and risk assessment) and calls the corresponding agents;

[0089] (3). Each agent acquires rules / knowledge from the knowledge and memory layer, and combines them with the model / engine of the reasoning and model layers to generate results;

[0090] (4) Conflicts in the detection results of the intelligent agent collaboration layer (such as contradictions between risk control and compliance conclusions) are resolved through the strategy of "priority-evidence verification-manual review";

[0091] (5) Aggregate the results and match them with the decision template to generate the final response / approval conclusion, and at the same time generate the decision evidence chain.

[0092] Optionally, in the financial business processing method provided in the embodiments of this application, the financial intelligent system includes: a knowledge memory layer, which stores processing rules for processing financial business, and based on S intelligent agents associated with sub-tasks, processes business for a second object to obtain business processing results, including: based on the intelligent agents associated with the S sub-tasks, calling the processing rules in the knowledge memory layer to process the intermediate representation to obtain S processing results; prioritizing the S processing results to obtain S sorted processing results; and generating a business processing result based on the sorted S processing results and the target language model.

[0093] Processing rules can be structured sets of instructions that define financial business logic, compliance requirements, and decision-making conditions, such as risk control thresholds or compliance clauses. The S processing results can be preliminary decision or judgment information output by S intelligent agents based on the IR and the invoked rules.

[0094] In one optional example, the collaboration layer can distribute S subtasks to corresponding agents. Each agent can retrieve processing rules related to the current subtask from the knowledge memory layer, combine them with entity and contextual information in the intermediate representation to perform reasoning, and generate S processing results. Then, the collaboration layer performs conflict detection and priority ranking on the S processing results. For example, if risk control recommends approval but compliance indicates missing materials, the compliance conclusion is placed first according to the "compliance priority" strategy. The ranked result sequence can be used as key constraints input to the target language model. The target language model integrates the hard constraints of high-priority results with auxiliary information from other results, performs semantic fusion and logical verification, and finally generates a business processing result (such as explicitly rejecting and explaining the reasons, or suggesting retrying after supplementing materials), ensuring the accuracy and interpretability of the decision.

[0095] Optionally, in the financial business processing method provided in this application embodiment, after generating the business processing result based on the sorted S processing results, the method further includes: generating reply content based on a preset template when the business processing result indicates that preset information needs to be supplemented; and reviewing the content supplemented by the second object after the second object has supplemented the preset information based on the reply content.

[0096] Preset information can refer to essential data items (such as income verification or identity verification) that are explicitly stated, missing, or require verification in the business processing results. Preset templates can be predefined standardized response formats used to generate clear and user-friendly prompts, ensuring consistency in information delivery. "Review" can be a secondary verification process for the legality, completeness, and authenticity of the supplementary content from the second party; for example, it can be completed by combining a rules engine or manual review.

[0097] In one optional example, when the business processing result indicates "preset information needs to be supplemented," the system can prompt the second party with the specific missing information and how to supplement it, and provide feedback to the second party through the multimodal access layer. After the second party completes the information supplementation via the APP or other channels, the review process can be triggered. The multimodal access layer can convert the newly added content back into an intermediate representation, and the collaboration layer merges it with the original IR, and re-invokes the relevant intelligent agent (such as the compliance intelligent agent) to conduct a special review of the supplemented content. If the review is successful, the subsequent business logic continues to execute; if the review fails or new problems are found, supplementary guidance is generated again or the process is terminated, forming a closed-loop processing mechanism of "prompt-supplement-review" to ensure the integrity and compliance of business data.

[0098] Optionally, in the financial business processing method provided in the embodiments of this application, after the intelligent agents associated with S sub-tasks process the business for the second object and obtain the business processing result, the method further includes: obtaining the call records and rule call data of the intelligent agents associated with the S sub-tasks, wherein the rule call data includes: data of the rules called by the intelligent agents associated with the S sub-tasks from the knowledge memory layer; and generating an evidence chain based on the call records, rule call data, and intermediate representation.

[0099] The call log can include: timestamps of the agent's task execution, input / output, and status logs. Rule call data can refer to the specific rule ID, version, and triggering conditions retrieved by the agent from the knowledge memory layer. The evidence chain is a traceable data package that records the decision-making basis throughout the entire process from input to output, used for auditing and compliance verification.

[0100] In one optional example, after a business transaction is completed, the call records of each agent and the corresponding rule call data, including the specific rule version referenced, can be collected. This dynamically executed data can then be correlated and integrated with a static intermediate representation (IR). Through timestamp alignment and context mapping, a complete chain of evidence is constructed, including the input IR, agent behavior, rule triggering, and model version. This chain of evidence can be encrypted and stored in the audit log, supporting not only real-time traceability and interpretation of any business decision but also meeting the stringent requirements of financial regulation for "decision traceability," ensuring the transparency and security of business processing, and achieving end-to-end auditability.

[0101] The processing flow of the financial business processing method provided in this embodiment will be described below in conjunction with a specific business scenario:

[0102] Scenario 1: Personal consumer loan application:

[0103] Figure 5 This is a timing diagram of loan application intelligent agent collaboration provided in the embodiments of this application. The following is in conjunction with... Figure 5 Explanation:

[0104] (1) Input and IR generation:

[0105] The customer (corresponding to the second party) submits a loan application form (including information such as loan amount of 500,000 yuan, personal credit score of 700, and proof of income) through the financial institution's APP.

[0106] The multimodal access layer is converted to IR.

[0107] (2) Task splitting and invocation of intelligent agents:

[0108] The agent collaboration layer is broken down into three sub-tasks:

[0109] Subtask 1 (Authentication) → Customer Service AI Agent;

[0110] Subtask 2 (Risk Assessment) → Risk Control Intelligent Agent;

[0111] Subtask 3 (Compliance Check) → Compliance Agent.

[0112] (3) Each agent accesses knowledge and memory layer resources:

[0113] The customer service AI agent verifies the customer's identity;

[0114] The risk control AI calls the "Personal Consumer Loan Risk Control Rules" (credit_score ≥ 650 and loan_amount ≤ 500,000 → low risk);

[0115] The compliant AI agent invoked the "Personal Consumer Loan Material Requirements" clause.

[0116] (4) Result conflict detection and resolution:

[0117] The agent returns the following results:

[0118] Customer service: "Identity verification passed";

[0119] Risk control: "Low risk level, approval recommended";

[0120] Compliance: "Income verification for the past 3 months is missing; it is recommended to provide it."

[0121] The collaboration layer triggers a priority strategy (compliance has a higher priority), and the result is "supplementary income proof is required".

[0122] (5) Decision generation and evidence chain recording:

[0123] The template generated a reply: "Your loan application requires supplementary income proof for the past 3 months. The review will continue after you provide the proof."

[0124] The audit layer generates a chain of evidence, including input IR, agent call records, and rule reference IDs (risk control rule R20260105-001, compliance clause C202503-002).

[0125] (6) Update intervention (if necessary):

[0126] If the business personnel (corresponding to the first object) discover that the new business processing of "credit score ≥ 700 exempts income proof" has been omitted, submit a correction and update request:

[0127] After sandbox fine-tuning and A / B testing, the updated requirements will be rolled out in a phased manner, and the new rules will be automatically applied to subsequent similar applications.

[0128] Scenario 2: Intelligent customer service for early repayment consultation:

[0129] (1) Input and IR generation:

[0130] A customer sent a message through the app's customer service: "I want to prepay my mortgage. What are the requirements?"

[0131] The multimodal access layer is converted to IR.

[0132] (2) Agent invocation and reasoning:

[0133] The intelligent agent collaboration layer assigns tasks to customer service intelligent agents;

[0134] The customer service AI retrieves the "mortgage prepayment rules" from the knowledge and memory layer (1 year of repayment + 15 days in advance application), and generates a reply by combining the inference layer LLM (Large Language Model, corresponding to the target language model): "You need to meet the requirement of 1 year of mortgage repayment and submit an application through the APP 15 days in advance."

[0135] (3) Update and optimization:

[0136] If the latest business processing rule is changed to "apply 10 days in advance", business personnel should submit an example update request:

[0137] After the update request was fine-tuned in the sandbox, the customer service AI agent's subsequent replies were automatically updated to "Apply 10 days in advance".

[0138] Through this embodiment, business personnel in financial institutions can directly participate in AI optimization, shortening the demand implementation cycle from "weekly" to "hourly"; achieving full-chain traceability through the decision evidence chain meets financial regulatory requirements, reducing audit manpower costs by 60%, thus achieving the goal of reducing audit costs; unified storage of rules / knowledge in the knowledge and memory layers allows customer service, risk control, and other systems to share data, reducing redundant development costs by 50% and enhancing cross-scenario reusability; the sandbox model + canary release mechanism avoids direct changes to the production environment, reducing the model anomaly rate from 3% to below 0.5%, thereby improving the stability of the financial intelligent system.

[0139] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0140] Example 2

[0141] This application also provides a financial transaction processing apparatus. It should be noted that the financial transaction processing apparatus of this application can be used to execute the financial transaction processing method provided in this application. The following describes the financial transaction processing apparatus provided in this application.

[0142] According to an embodiment of this application, an apparatus for implementing the above-described financial transaction processing method is also provided, such as... Figure 6 As shown, the device includes a receiving unit 61, a first processing unit 62, and a second processing unit 63.

[0143] The receiving unit 61 is used to receive the update request submitted by the first object, wherein the first object includes: an object in a financial institution that handles financial business for customers, and the update request includes: a request to update the business processing function of the financial intelligent system, wherein the financial intelligent system includes: M intelligent agents for processing financial business and a target language model, where M is a positive integer;

[0144] The first processing unit 62 is used to parse the update requirements, obtain the parsing results, and generate training tasks based on the parsing results;

[0145] The second processing unit 63 is used to update the financial intelligence system based on the training task using a sandbox model to obtain the updated financial intelligence system. The sandbox model includes a copy of the already deployed model in the financial intelligence system that is isolated from the production environment of the financial intelligence system.

[0146] In the financial business processing apparatus provided in this application embodiment, the receiving unit 61 can receive update requests submitted by a first object, wherein the first object includes an object in a financial institution that handles financial business for customers, and the update request includes a request to update the business processing functions of the financial intelligent system. The financial intelligent system includes M intelligent agents for processing financial business and a target language model, where M is a positive integer. The first processing unit 62 parses the update request to obtain the parsing result, and generates a training task based on the parsing result. The second processing unit 63 updates the financial intelligent system based on the training task using a sandbox model to obtain the updated financial intelligent system. The sandbox model includes a copy of the already deployed model in the financial intelligent system that is isolated from the production environment of the financial intelligent system. This solves the technical problem in related technologies where the development team has to develop and implement update requests in the financial business system, resulting in low efficiency in implementing new update requests. In this embodiment, based on the update request input by the business handling object, the model in the financial intelligent system is adjusted using a sandbox model, avoiding the situation in related technologies where the development team has to develop and implement update requests, resulting in low efficiency. This achieves the technical effect of improving the efficiency of developing and implementing business requirements for the financial system.

[0147] Optionally, in the financial business processing apparatus provided in this application embodiment, the second processing unit includes: a training subunit, used to train the sandbox model based on a training task until the sandbox model meets a preset convergence condition, thereby obtaining the trained sandbox model; a testing subunit, used to test the financial intelligence system and the trained sandbox model using an A / B testing strategy, thereby obtaining test results; and an updating subunit, used to update the financial intelligence system based on the trained sandbox model when the test results indicate that the accuracy of the trained sandbox model in processing financial business is higher than that of the financial intelligence system.

[0148] Optionally, in the financial business processing apparatus provided in the embodiments of this application, the financial business processing apparatus further includes: a publishing unit, used to perform gray-scale release of the updated financial intelligent system based on a preset ratio after updating the financial intelligent system using a sandbox model based on a training task to obtain the updated financial intelligent system; and a recording unit, used to record a snapshot of the updated financial system and the update requirements to an audit log after the gray-scale release of the updated financial intelligent system is completed.

[0149] Optionally, in the financial business processing apparatus provided in the embodiments of this application, the financial business processing apparatus further includes: a receiving unit, configured to receive a business processing request submitted by a second object after updating the financial intelligent system based on a training task and using a sandbox model to obtain an updated financial intelligent system, and to perform structured processing on the business processing request to obtain an intermediate representation, wherein the second object includes: an object that receives services from a financial institution; a determining unit, configured to determine S sub-tasks based on the intermediate representation, wherein each sub-task is associated with one intelligent agent, and S is a positive integer less than M; and a third processing unit, configured to process the business for the second object based on the intelligent agents associated with the S sub-tasks, and obtain a business processing result.

[0150] Optionally, in the financial business processing apparatus provided in this application embodiment, the financial intelligent system includes: a knowledge memory layer, which stores processing rules for processing financial business; and a third processing unit, which includes: a processing subunit, used to call the processing rules in the knowledge memory layer based on S intelligent agents associated with subtasks, to process intermediate representations and obtain S processing results; a sorting subunit, used to prioritize the S processing results and obtain S sorted processing results; and a generation subunit, used to generate business processing results based on the S sorted processing results and a target language model.

[0151] Optionally, in the financial business processing apparatus provided in this application embodiment, the third processing unit further includes: a generation subunit, used to generate a response content based on a preset template after generating a business processing result based on the sorted S processing results, when the business processing result indicates that preset information needs to be supplemented; and an auditing subunit, used to audit the content supplemented by the second object after the second object has supplemented the preset information based on the response content.

[0152] Optionally, in the financial business processing apparatus provided in the embodiments of this application, the financial business processing apparatus further includes: an acquisition unit, configured to acquire the call records and rule call data of the intelligent agents associated with the S sub-tasks after the intelligent agents associated with the S sub-tasks have processed the business for the second object and obtained the business processing result, wherein the rule call data includes: data of the rules called by the intelligent agents associated with the S sub-tasks from the knowledge memory layer; and a generation unit, configured to generate an evidence chain based on the call records, rule call data, and intermediate representation.

[0153] It should be noted that the receiving unit 61, the first processing unit 62, and the second processing unit 63 mentioned above correspond to steps S201 to S203 in Embodiment 1. Each unit and the corresponding step implement the same instance and application scenario, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules or units can be hardware or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above modules can also be part of the device and run in the computer terminal 10 provided in Embodiment 1.

[0154] Example 3

[0155] Embodiments of this application may provide an electronic device. Figure 7 This is a structural block diagram of an electronic device according to an embodiment of this application. Figure 7 As shown, the electronic device may include: one or more ( Figure 7 (Only one is shown) Processor 702, memory 704, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0156] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and apparatus in the embodiments of this application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the above-described methods. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0157] The processor can invoke information and applications stored in the memory via a transmission device to perform the following steps: receiving an update request submitted by a first object, wherein the first object includes an object in a financial institution that handles financial business for customers, and the update request includes a request to update the business processing functions of the financial intelligence system, wherein the financial intelligence system includes M intelligent agents for processing financial business and a target language model, where M is a positive integer; parsing the update request to obtain the parsing result, and generating a training task based on the parsing result; updating the financial intelligence system using a sandbox model based on the training task to obtain the updated financial intelligence system, wherein the sandbox model includes a copy of the already deployed model in the financial intelligence system that is isolated from the production environment of the financial intelligence system.

[0158] The processor can also access information and applications stored in the memory via a transmission device to execute the following steps: updating the financial intelligence system based on the training task using a sandbox model to obtain an updated financial intelligence system, including: training the sandbox model based on the training task until the sandbox model meets a preset convergence condition to obtain a trained sandbox model; testing the financial intelligence system and the trained sandbox model using an A / B testing strategy to obtain test results; and updating the financial intelligence system based on the trained sandbox model if the test results indicate that the accuracy of the trained sandbox model in processing financial transactions is higher than that of the financial intelligence system.

[0159] The processor can also access information and applications stored in the memory via a transmission device to perform the following steps: after updating the financial intelligence system using a sandbox model based on the training task to obtain the updated financial intelligence system, the processor can also perform a canary release of the updated financial intelligence system based on a preset ratio; and after the canary release of the updated financial intelligence system is completed, a snapshot of the updated financial system and the update requirements are recorded in the audit log.

[0160] The processor can also access information and applications stored in the memory via a transmission device to execute the following steps: After updating the financial intelligence system using a sandbox model based on the training task to obtain the updated financial intelligence system, the process further includes: receiving a business processing request submitted by a second object and performing structured processing on the business processing request to obtain an intermediate representation, wherein the second object includes an object that receives services from a financial institution; determining S sub-tasks based on the intermediate representation, wherein each sub-task is associated with one intelligent agent, and S is a positive integer less than M; and processing business for the second object based on the intelligent agents associated with the S sub-tasks to obtain the business processing result.

[0161] The processor can also invoke information and applications stored in the memory via a transmission device to execute the following steps: The financial intelligence system includes: a knowledge memory layer, which stores processing rules for financial transactions, and intelligent agents associated with S sub-tasks to process transactions for a second object and obtain transaction results, including: based on the intelligent agents associated with S sub-tasks, invoking the processing rules in the knowledge memory layer to process intermediate representations and obtain S processing results; prioritizing the S processing results to obtain S sorted processing results; and generating a transaction result based on the sorted S processing results and the target language model.

[0162] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: after generating a business processing result based on the sorted S processing results, it also includes: generating reply content based on a preset template when the business processing result indicates that preset information needs to be supplemented; and reviewing the content supplemented by the second object after the second object has supplemented the preset information based on the reply content.

[0163] The processor can also invoke information and applications stored in the memory via a transmission device to perform the following steps: after the intelligent agents associated with S subtasks handle business for the second object and obtain the business handling result, it further includes: obtaining the invocation records and rule invocation data of the intelligent agents associated with the S subtasks, wherein the rule invocation data includes: data of the rules invoked by the intelligent agents associated with the S subtasks from the knowledge memory layer; and generating an evidence chain based on the invocation records, rule invocation data, and intermediate representation.

[0164] By adopting the embodiments of this application, based on the update requirements input by the business processing object, the model in the financial intelligent system is adjusted through a sandbox model, avoiding the situation in related technologies where the development team needs to develop and implement the update requirements, which is inefficient. This achieves the technical effect of improving the efficiency of developing and implementing the business requirements of the financial system.

[0165] Those skilled in the art will understand that Figure 7 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 7 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 7 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 7 The different configurations shown.

[0166] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0167] Example 4

[0168] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the financial transaction processing method provided in Embodiment 1.

[0169] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0170] This application also provides a computer program product that, when executed on a data processing device, is suitable for performing steps of a financial business processing method.

[0171] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0172] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0173] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0174] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0175] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0176] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0177] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A financial transaction processing method, characterized in that, include: The system receives an update request submitted by a first object, wherein the first object includes an object in a financial institution that handles financial business for customers, and the update request includes a request to update the business processing functions of the financial intelligent system, wherein the financial intelligent system includes M intelligent agents for processing the financial business and a target language model, where M is a positive integer; The update requirement is parsed to obtain the parsing result, and a training task is generated based on the parsing result; Based on the training task, the financial intelligence system is updated using a sandbox model to obtain an updated financial intelligence system. The sandbox model includes a copy of the already deployed model in the financial intelligence system that is isolated from the production environment of the financial intelligence system.

2. The processing method according to claim 1, characterized in that, Based on the training task, the financial intelligence system is updated using a sandbox model to obtain the updated financial intelligence system, including: Based on the training task, the sandbox model is trained until the sandbox model meets the preset convergence condition, and the trained sandbox model is obtained. The financial intelligence system and the trained sandbox model were tested using an A / B testing strategy, and the test results were obtained. If the test results indicate that the accuracy of the trained sandbox model in processing the financial business is higher than that of the financial intelligence system, the financial intelligence system is updated based on the trained sandbox model.

3. The processing method according to claim 1, characterized in that, After updating the financial intelligence system using a sandbox model based on the training task to obtain the updated financial intelligence system, the process further includes: The updated financial intelligence system will be released in a gray-scale manner based on a preset ratio. Once the updated financial intelligence system has been released in a phased rollout, a snapshot of the updated financial system and the update request will be recorded in the audit log.

4. The processing method according to claim 1, characterized in that, After updating the financial intelligence system using a sandbox model based on the training task to obtain the updated financial intelligence system, the process further includes: The system receives a business processing request submitted by a second object and performs structured processing on the business processing request to obtain an intermediate representation. The second object includes an object that receives services from the financial institution. Based on the intermediate representation, S subtasks are determined, wherein each subtask is associated with one of the agents, and S is a positive integer less than M; Based on the S intelligent agents associated with the sub-tasks, the second object is processed for business, and the business processing result is obtained.

5. The processing method according to claim 4, characterized in that, The financial intelligence system includes: a knowledge memory layer, which stores processing rules for the financial transactions, and processes transactions for the second object based on S intelligent agents associated with the sub-tasks, obtaining transaction results, including: Based on the S agents associated with the subtasks, the processing rules in the knowledge memory layer are invoked to process the intermediate representation, resulting in S processing results. The S processing results are sorted by priority to obtain S sorted processing results; Based on the sorted S processing results and the target language model, a business processing result is generated.

6. The processing method according to claim 5, characterized in that, After generating the business processing result based on the S sorted processing results and the target language model, the process further includes: If the business processing result indicates that preset information needs to be supplemented, a reply content is generated based on a preset template; After the second object has completed supplementing the preset information based on the reply content, the supplemented content of the second object is reviewed.

7. The processing method according to claim 5, characterized in that, After the intelligent agents associated with the S sub-tasks process the business for the second object and obtain the business processing result, the process further includes: Obtain the call records and rule call data of the S agents associated with the subtasks, wherein the rule call data includes: data of the rules called from the knowledge memory layer by the S agents associated with the subtasks; An evidence chain is generated based on the call records, the rule call data, and the intermediate representation.

8. A financial transaction processing device, characterized in that, include: A receiving unit is configured to receive an update request submitted by a first object, wherein the first object includes an object in a financial institution that handles financial business for customers, and the update request includes a request to update the business processing function of a financial intelligent system, wherein the financial intelligent system includes M intelligent agents and a target language model for processing the financial business, where M is a positive integer; The first processing unit is used to parse the update request, obtain the parsing result, and generate a training task based on the parsing result; The second processing unit is used to update the financial intelligence system based on the training task using a sandbox model to obtain an updated financial intelligence system, wherein the sandbox model includes a copy of the already deployed model in the financial intelligence system that is isolated from the production environment of the financial intelligence system.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the computer-readable storage medium is located to perform the financial business processing method according to any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the financial business processing method according to any one of claims 1 to 7.

Images