Intelligent agent-based financial service processing method, device, equipment and storage medium
The intelligent agent processing solution, which employs dual-channel stream processing and candidate action detection, addresses issues such as user cognitive bias, unauthorized operations, and privacy leaks in financial transactions, achieving highly secure and accurate financial transaction processing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- FOUNDER SECURITIES CO LTD
- Filing Date
- 2026-06-03
- Publication Date
- 2026-07-31
AI Technical Summary
Existing intelligent agent processing solutions in the financial field suffer from problems such as user cognitive bias, unauthorized business operations, accidental data modification, privacy leaks, and difficulty in fault tracing, making it difficult to meet the high security, high accuracy, and high controllability requirements of financial businesses.
A dual-channel stream processing method is adopted, which generates the execution context through state machine initialization, separates the model reasoning and display content, performs candidate action detection and verification, and generates processing results by combining structured financial data, thereby achieving accurate intelligent agent processing.
It improves the security and reliability of financial services, reduces user cognitive bias, prevents unauthorized operations, ensures data privacy protection, and improves troubleshooting efficiency.
Smart Images
Figure CN122332665B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of finance and securities, and in particular to a method, apparatus, device, and storage medium for financial business processing based on intelligent agents. Background Technology
[0002] With the rapid development of large-scale model technology, intelligent agents have been widely deployed in financial and securities web business scenarios. Relying on the understanding and generation capabilities of large language models, they enable intelligent services such as intelligent customer service and business assistance. At the same time, front-end computing power continues to upgrade, and the downward shift of computing power has become an industry trend. However, existing intelligent agent processing solutions in the financial field still have obvious technical defects and are difficult to adapt to the core requirements of high security, high accuracy, and high controllability in financial business.
[0003] Currently, most mainstream intelligent agent architectures in the industry adopt a server-side centralized orchestration or a simple client-side API (Application Programming Interface) call mode. The basic architecture only carries a standard ReAct (Reasoning and Acting) execution flow, resulting in a rudimentary overall technical architecture design and a severe lack of control mechanisms. First, there is no channel isolation between the model's inference content and the content displayed to the user, allowing for arbitrary rendering and display of secondary target fragments, which can easily lead to user cognitive biases and violate compliance red lines in the financial industry. Second, action execution relies entirely on the model's autonomous judgment, lacking standardized pre-gating checks, making it prone to unauthorized business operations and erroneous data modification, directly threatening the security of financial data and business operations. Furthermore, existing solutions have simplistic exception handling logic, relying solely on program errors as the basis for retry decisions, without constraints on the number of retries or runtime status. This easily leads to infinite loops and high-frequency interface requests, causing page lag, interface congestion, and in severe cases, service avalanche. In addition, the business interaction context is directly injected into the model without sensitive information clipping and desensitization, which poses a risk of privacy leakage. Furthermore, the lack of execution logs and rollback records throughout the process makes it impossible to trace and review faults. Coupled with the problem of the disconnect between the front-end and back-end development workflow, this further reduces the efficiency of iteration and troubleshooting of financial intelligent applications.
[0004] Therefore, designing a precise intelligent agent processing solution to meet the operational needs of the high-risk financial scenario is a pressing technical problem that needs to be solved. Summary of the Invention
[0005] In view of this, the purpose of this invention is to provide a method, apparatus, device, and storage medium for financial business processing based on intelligent agents, capable of designing a precise intelligent agent processing solution to meet the operational needs of the high-risk financial scenario. The specific solution is as follows: Firstly, this application provides a financial transaction processing method based on an intelligent agent, applied to a financial transaction processing intelligent agent on a web-based financial transaction platform, comprising: Based on the acquired financial business request, the state machine performs corresponding initialization operations to generate an execution context based on the financial business request and the initialization operations, and uses the financial business request and the execution context to generate financial context data; the state machine is a shared state machine of the financial business processing agent and the financial business web client. Based on the financial context data, a corresponding streaming financial segment is determined, and a first target segment in the streaming financial segment is buffered to a preset display channel, while a second target segment in the streaming financial segment is buffered to a preset inference channel; the first target segment is a financial segment that can be directly displayed on the financial business web terminal; the second target segment is a financial segment that needs to be decided. In the preset reasoning channel, candidate actions are determined based on the second target fragment, and the candidate actions are detected to generate corresponding structured financial data based on the detected candidate actions. Based on the structured financial data and the first target fragment in the preset display channel, a financial business processing result corresponding to the financial business request is generated.
[0006] Optionally, the step of performing corresponding initialization operations on the state machine based on the acquired financial business request to generate an execution context based on the financial business request and the initialization operations includes: Obtain financial service requests and create corresponding session contexts based on the financial service requests; The current state of the state machine in the web client of the financial business is initialized to a preset thinking state, and the retry budget and preset log cache corresponding to the financial business request are initialized to complete the corresponding initialization operations. An execution context is generated based on the financial transaction request, the session context, and the initialization operation.
[0007] Optionally, generating financial context data using the financial transaction request and the execution context includes: Determine the current page's Uniform Resource Locator, routing information, component status, and user selection information to determine the first target data for the financial business web application; The first target data is cropped based on preset data relevance conditions to obtain the second target data; the preset data relevance conditions are the conditions for judging the relevance between the first target data and the financial business request. Based on preset field sensitivity conditions, target sensitive fields are determined from the second target data, and preset desensitization and preset prompt word injection operations are performed on the target sensitive fields in the second target data to obtain the corresponding third target data; Financial context data is generated based on the third target data and the execution context.
[0008] Optionally, determining the corresponding streaming financial segment based on the financial context data includes: Based on the financial context data, a target request is generated that is adapted to the large language model in the financial business processing agent, and a streaming channel is created. The streaming channel is used to receive streaming financial segments generated by the large language model based on the target request.
[0009] Optionally, the step of detecting the candidate actions to generate corresponding structured financial data based on the detected candidate actions includes: The skill identifier, parameter information, and target object in the candidate actions are determined to identify the action object of the candidate actions. Identify the target skill corresponding to the skill identifier, and determine whether there is a skill execution condition corresponding to the target skill on the financial business web terminal, so as to complete the first detection operation of the candidate action; Determine whether the parameter information satisfies the preset skill declaration structure to complete the second detection operation of the candidate action; Determine whether the risk level of the candidate action meets the preset risk range, so as to complete the third detection operation of the candidate action; Based on the first detection operation, the second detection operation, and the third detection operation, the candidate actions that pass the detection are determined, so as to generate corresponding structured financial data by the skill executor in the financial business web terminal based on the action object and the candidate actions that pass the detection.
[0010] Optionally, after detecting the candidate action, the method further includes: If the candidate action fails the detection, the current state of the state machine is set to the preset rollback state, and the retry budget is deducted to determine the retry budget after deduction. If the deducted retry budget meets the preset availability conditions, then the current state of the state machine is set to the preset thinking state.
[0011] Optionally, generating the financial transaction processing result corresponding to the financial transaction request based on the structured financial data and the first target fragment in the preset display channel includes: The first target segment in the preset display channel is subjected to integrity and legality checks, and the first target segment that passes the checks is subjected to preset rendering operations to obtain the corresponding rendering data. The financial transaction processing result corresponding to the financial transaction request is determined based on the rendered data and the structured financial data.
[0012] Secondly, this application provides a financial transaction processing device based on an intelligent agent, which is applied to a financial transaction processing intelligent agent on a financial transaction web interface, comprising: The first data generation module is used to perform corresponding initialization operations on the state machine based on the acquired financial business request, so as to generate an execution context based on the financial business request and the initialization operation, and generate financial context data using the financial business request and the execution context; the state machine is a state machine shared by the financial business processing intelligent agent and the financial business web client; The segment buffering module is used to determine the corresponding streaming financial segments based on the financial context data, buffer the first target segment in the streaming financial segment to a preset display channel, and buffer the second target segment in the streaming financial segment to a preset inference channel; the first target segment is a financial segment that can be directly displayed on the financial business web terminal; the second target segment is a financial segment that needs to be decided. The second data generation module is used to determine candidate actions based on the second target fragment in the preset inference channel, and to detect the candidate actions in order to generate corresponding structured financial data based on the detected candidate actions. The result generation module is used to generate a financial business processing result corresponding to the financial business request based on the structured financial data and the first target fragment in the preset display channel.
[0013] Thirdly, this application provides an electronic device, comprising: Memory, used to store computer programs; A processor is used to execute the computer program to implement the aforementioned agent-based financial business processing method.
[0014] Fourthly, this application provides a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned agent-based financial business processing method.
[0015] In this application, a state machine is initialized based on the acquired financial business request to generate an execution context, and financial context data is generated using the financial business request and the execution context. The state machine is shared by the financial business processing agent and the financial business web client. Based on the financial context data, a corresponding streaming financial segment is determined. A first target segment in the streaming financial segment is buffered to a preset display channel, and a second target segment in the streaming financial segment is buffered to a preset inference channel. The first target segment is a financial segment that can be directly displayed on the financial business web client; the second target segment is a financial segment that needs to be decided. In the preset inference channel, candidate actions are determined based on the second target segment, and the candidate actions are detected to generate corresponding structured financial data based on the detected candidate actions. A financial business processing result corresponding to the financial business request is generated based on the structured financial data and the first target segment in the preset display channel. As can be seen from the above, in this application, the shared state machine is initialized based on the acquired financial business request. An execution context is constructed based on the financial business request and the initialization operation, and then financial context data is generated by combining the financial business request and the execution context. The corresponding streaming financial segments are matched based on the financial context data. The first target segment, which can be directly presented on the financial business web interface, is cached in a preset display channel, while the second target segment, which requires decision processing, is cached in a preset inference channel. Within the preset inference channel, candidate actions are derived based on the second target segment, and these candidate actions are validated and detected. Qualified candidate actions generate corresponding structured financial data. Finally, the structured financial data is combined with the first target segment in the preset display channel to generate a financial business processing result matching the financial business request. In this way, this application can design a precise intelligent agent processing solution to meet the operational needs of high-risk financial scenarios. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0017] Figure 1 This is a flowchart of a financial business processing method based on intelligent agents disclosed in this application; Figure 2 This is a schematic diagram of a layered and gated structure disclosed in this application; Figure 3 This application discloses a state machine flowchart; Figure 4 This is a failure path diagram disclosed in this application; Figure 5 This is a schematic diagram of the structure of a financial transaction processing device based on an intelligent agent disclosed in this application; Figure 6 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] Currently, most mainstream intelligent agent architectures adopt a server-side centralized orchestration or a simple client-side API call model, with the basic architecture only equipped with a conventional ReAct execution process. The overall technical architecture design is rudimentary, and the control mechanism is severely lacking. First, the model's inference content and the content displayed to the user are not isolated from each other, and the second target fragment is rendered and displayed arbitrarily, which can easily lead to user cognitive bias and cross the compliance red line of the financial industry. Second, the execution of actions relies entirely on the model's autonomous judgment, lacking standardized pre-gating verification, which can easily lead to unauthorized business operations, data modification errors, and other problems, directly threatening the security of financial data and business operations. Third, the existing solutions have a simple exception handling logic, using program errors as the basis for retry judgment, without constraints on the number of retries and running status. This can easily lead to infinite loops and high-frequency interface requests, causing page lag, interface congestion, and in severe cases, service avalanche. In addition, the business interaction context is directly injected into the model without sensitive information trimming and de-identification, posing a risk of privacy leakage. Moreover, the lack of execution logs and rollback records makes it impossible to trace and review faults. Coupled with the problem of the disconnect between the front-end and back-end development workflow, this further reduces the iteration and troubleshooting efficiency of financial intelligent applications. To this end, this application provides a method, apparatus, device and storage medium for financial business processing based on intelligent agents, which can design a precise intelligent agent processing solution to meet the operational needs of the high-risk financial scenario.
[0020] See Figure 1 As shown, this embodiment of the invention discloses a financial business processing method based on an intelligent agent, applied to a financial business processing intelligent agent on a financial business web interface, comprising: Step S11: Perform corresponding initialization operations on the state machine based on the acquired financial business request, so as to generate an execution context based on the financial business request and the initialization operations, and generate financial context data using the financial business request and the execution context; the state machine is a state machine shared by the financial business processing agent and the financial business web client.
[0021] In this embodiment, the financial transaction processing agent acquires a financial transaction request and creates a corresponding session context based on the request. This session context records temporary data and intermediate states generated during the current financial transaction processing. The financial transaction request can be triggered by a user on the front-end interface of the financial transaction web application. It should be noted that the financial transaction processing agent can acquire the financial transaction request in various ways. In one specific implementation, the agent can capture request data generated by user interaction by listening to the event bus of the financial transaction web application. In another specific implementation, the financial transaction web application can proactively construct a financial transaction request after receiving a user operation and send it to the agent through a predefined application programming interface (API).
[0022] In this embodiment, it is also necessary to initialize the current state of the state machine in the financial business web client, and at the same time, it is also necessary to initialize the retry budget and preset log cache corresponding to the financial business request. Specifically, the current state of the state machine in the financial business web client is initialized to a preset thinking state, and the retry budget and preset log cache corresponding to the financial business request are initialized to complete the corresponding initialization operations.
[0023] After completing the above initialization operations, an execution context is generated based on the financial business request, the session context, and the initialization operations. The execution context can be a container object that aggregates all the information required for the current processing, so as to quickly read and update the required data at different stages of the financial business processing.
[0024] Furthermore, financial context data is generated using the financial business request and execution context. Specifically, the current page Uniform Resource Locator (URL), routing information, component status, and user selection information are determined to identify the first target data for the financial business web application. Based on preset data relevance conditions, the first target data is pruned to obtain second target data. These preset data relevance conditions are used to determine the relevance between the first target data and the financial business request. Based on preset field sensitivity conditions, target sensitive fields are identified from the second target data, and preset desensitization and preset prompt word injection operations are performed on these target sensitive fields to obtain corresponding third target data. Financial context data is generated based on the third target data and the execution context. This financial context data will serve as the basis for decision-making and action execution by the financial business processing agent in subsequent steps.
[0025] Step S12: Based on the financial context data, determine the corresponding streaming financial segment, buffer the first target segment in the streaming financial segment to a preset display channel, and buffer the second target segment in the streaming financial segment to a preset inference channel; the first target segment is a financial segment that can be directly displayed on the financial business web terminal; the second target segment is a financial segment that needs to be decided.
[0026] In this embodiment, after generating financial context data, the financial business processing agent generates a target request adapted to the large language model within the agent based on the financial context data, and creates a streaming channel. This streaming channel can be used to establish a continuous data transmission connection with the large language model. In one specific implementation, the streaming channel can be implemented via server push events. Then, the streaming channel is used to receive streaming financial segments generated by the large language model based on the target request.
[0027] The aforementioned streaming financial segments can be response content output segment by segment by a large language model. Each segment can be a word, a sentence, or a complete financial operation instruction. Upon receiving the streaming financial segments, the financial business processing agent can perform type identification on each segment, identifying the financial segments that can be directly displayed on the financial business web interface as the first target segments and buffering them to a preset display channel; identifying the financial segments that require decision-making as the second target segments and buffering them to a preset inference channel. The preset display channel is connected to the front-end rendering component of the financial business web interface, and the preset inference channel can be connected to the decision engine inside the financial business processing agent.
[0028] Step S13: In the preset inference channel, candidate actions are determined based on the second target fragment, and the candidate actions are detected to generate corresponding structured financial data based on the detected candidate actions.
[0029] In this embodiment, the financial business processing agent buffers the second target fragment to a preset inference channel and then processes the second target fragment within the preset inference channel. First, candidate actions are determined based on the second target fragment. These candidate actions can be structured instructions representing the intent of financial business operations in the output of a large language model. Next, the skill identifier, parameter information, and target object in the candidate actions are determined to identify the action object of the candidate actions.
[0030] After identifying candidate actions, multiple detection operations are performed on them. First, the target skill corresponding to the skill identifier is identified, and it is determined whether the financial business web terminal has a skill execution condition corresponding to the target skill, thus completing the first detection operation of the candidate action. Second, it is determined whether the parameter information meets the preset skill declaration structure, thus completing the second detection operation of the candidate action. Finally, it is determined whether the risk level of the candidate action meets the preset risk range, thus completing the third detection operation of the candidate action.
[0031] Furthermore, based on the first detection operation, the second detection operation, and the third detection operation, the candidate actions that pass the detection are determined, so that the skill executor in the financial business web client generates corresponding structured financial data based on the action object and the candidate actions that pass the detection. The aforementioned structured financial data can be standardized instructions that can be directly parsed and executed by the financial business web client.
[0032] In one specific implementation, if the candidate action detection fails, the current state of the state machine is set to a preset rollback state, and the retry budget is deducted to determine the deducted retry budget; if the deducted retry budget meets the preset availability conditions, the current state of the state machine is set to the preset thinking state, so that the financial business processing agent can trigger the large language model to regenerate the streaming financial segment again.
[0033] Step S14: Generate a financial business processing result corresponding to the financial business request based on the structured financial data and the first target fragment in the preset display channel.
[0034] In this embodiment, after obtaining the structured financial data, it is also necessary to process the first target segment in the preset display channel.
[0035] Specifically, the first target segment in the preset display channel undergoes integrity and legality checks, and the first target segment that passes the checks is subjected to preset rendering operations to obtain corresponding rendering data. Finally, based on the rendering data and the structured financial data, the financial business processing result corresponding to the financial business request is determined. In one specific embodiment, the rendering data can be used as the front-end display content, and the structured financial data can be used as the back-end execution result. Both can be encapsulated into a unified response object and returned to the financial business web client to complete the processing of this financial business request.
[0036] As can be seen from the above, in this application, the shared state machine is initialized based on the acquired financial business request. An execution context is constructed based on the financial business request and the initialization operation, and then financial context data is generated by combining the financial business request and the execution context. The corresponding streaming financial segments are matched based on the financial context data. The first target segment, which can be directly presented on the financial business web interface, is cached in a preset display channel, while the second target segment, which requires decision processing, is cached in a preset inference channel. Within the preset inference channel, candidate actions are derived based on the second target segment, and these candidate actions are validated and detected. Qualified candidate actions generate corresponding structured financial data. Finally, the structured financial data is combined with the first target segment in the preset display channel to generate a financial business processing result matching the financial business request. In this way, this application can design a precise intelligent agent processing solution to meet the operational needs of high-risk financial scenarios.
[0037] The following is combined with Figure 2 , Figure 3 , Figure 4 The schematic diagram shown illustrates the technical solution of the embodiments of this application in detail.
[0038] In web-based business within the financial securities sector, high demands are placed on the responsiveness of human-computer interaction, the accuracy of data processing, and the efficiency of problem troubleshooting. Agent-based intelligent systems can be applied to scenarios such as intelligent customer service, business consultation, and operational assistance. This solution proposes a "verifiable, constrainable, rollback-enabled, and auditable" edge-side ReAct processing method: namely, through methods such as... Figure 2 The dual-channel stream processing, three-stage skill gating, state machine rollback and restricted retry budget, and context-trimmed and desensitized injection demonstrated enable stable deployment of edge-side intelligent agents in high real-time and high-security scenarios. Among these features... Figure 2 This section primarily demonstrates the location of the dual-channel splitter unit and the three-stage gating unit within the four-layer architecture. Specifically, this solution is implemented according to the steps shown below.
[0039] Access request and state machine initialization. Input conditions: User submits a task request; Processing actions: Create a session context, initialize the state machine to the Thinking state (i.e., the default thinking state), initialize the retry budget and gating log cache; Output result: Obtain a controllable execution context. For details on the Thinking state, please refer to... Figure 3 As shown, Figure 3 It also includes a display of the transition conditions between Thinking, Acting (i.e., the execution state), Observing (i.e., the observation state), and Fallback (i.e., the default fallback state).
[0040] Context collection, trimming, and desensitization. Input conditions: Enter the pre-inference preparation stage; Processing actions: Collect page URLs (i.e., Uniform Resource Locators), routes, component states, selection information, etc.; Trim fields according to task relevance; Desensitize sensitive fields and inject prompt words; Output results: Obtain the secure context payload (i.e., financial context data) of the injectable model.
[0041] Model request sent and streaming connection established. Input conditions: Security context payload has been generated; Processing action: Send model request and establish streaming channel; Output result: Receive real-time output fragments from the model.
[0042] Dual-channel streaming splitting. Input condition: Streaming segments (i.e., streaming financial segments) arrive; Processing action: Split the segment into an inference channel and a display channel. The inference channel is only used for internal decision-making and does not enter the UI rendering chain; Output result: Forming an isolated dual-channel buffer.
[0043] Submit after display channel validation. Input conditions: The display channel has subscribing fragments; Processing actions: Perform structural integrity checks and content validity checks; Only fragments that pass the checks enter the rendering queue; Output results: Users only see subscribing content, avoiding inference pollution.
[0044] Action intent parsing. Input conditions: Action candidates are generated by the inference channel; Action processing: Skill identifiers, inputs, and target objects are parsed to generate candidate action descriptions; Output result: Action objects to be gated.
[0045] Three-stage skill gating. Input conditions: Candidate actions generated; Processing actions: First, scope matching, i.e., checking whether the current page, component, and route allow the skill (i.e., the first detection operation); Second, schema validation, i.e., checking whether the input parameters meet the skill declaration structure (i.e., the second detection operation); Third, risk permission, i.e., checking whether the risk level is within the current session's permitted range (i.e., the third detection operation). Output result: Output "Execution Allowed" or "Execution Denied + Reason Code".
[0046] Action execution and observation data collection. Input condition: Gating passed; Action processing: Invoking the skill executor, collecting execution results and state changes, and generating structured observations (i.e., structured financial data); Output result: Providing a basis for decision-making in the next round of the loop.
[0047] Failure rollback and budget deduction. Input conditions: gating failure, execution failure, or parsing failure; Processing actions: Enter Fallback state, record the reason for failure, and deduct the retry budget; if the budget is still available, return to Thinking, otherwise terminate; Output result: controlled failure processing chain. Furthermore, the failure path can be as follows... Figure 4 As shown.
[0048] Loop termination criteria. Input condition: Observation is available; Action: Determine if the task is accomplished; if accomplished, terminate and output the result; if not accomplished and the budget is sufficient, continue to the next round; Output result: Loop closed or enter the next round.
[0049] Accordingly, see Figure 5 As shown, this application provides an agent-based financial transaction processing device, which is applied to a financial transaction processing agent on a web-based financial transaction client, and includes: The first data generation module 11 is used to perform corresponding initialization operations on the state machine based on the acquired financial business request, so as to generate an execution context based on the financial business request and the initialization operation, and generate financial context data using the financial business request and the execution context; the state machine is a state machine shared by the financial business processing intelligent agent and the financial business web terminal. The segment buffer module 12 is used to determine the corresponding streaming financial segment based on the financial context data, buffer the first target segment in the streaming financial segment to a preset display channel, and buffer the second target segment in the streaming financial segment to a preset inference channel; the first target segment is a financial segment that can be directly displayed on the financial business web terminal; the second target segment is a financial segment that needs to be decided. The second data generation module 13 is used to determine candidate actions based on the second target fragment in the preset inference channel, and to detect the candidate actions so as to generate corresponding structured financial data based on the detected candidate actions. Result generation module 14 is used to generate a financial business processing result corresponding to the financial business request based on the structured financial data and the first target fragment in the preset display channel.
[0050] In some specific embodiments, the first data generation module 11 specifically includes: A context creation unit is used to obtain financial business requests and create corresponding session contexts based on the financial business requests; The information initialization unit is used to initialize the current state of the state machine in the financial business web client to a preset thinking state, and to initialize the retry budget and preset log cache corresponding to the financial business request, so as to complete the corresponding initialization operations. The context generation unit is used to generate an execution context based on the financial business request, the session context, and the initialization operation.
[0051] In some specific embodiments, the first data generation module 11 specifically includes: The first data determination unit is used to determine the current page Uniform Resource Locator, routing information, component status, and user selection information, so as to determine the first target data of the financial business web terminal; The second data determination unit is used to perform field trimming on the first target data based on preset data correlation conditions to obtain second target data; the preset data correlation conditions are the conditions for judging the correlation between the first target data and the financial business request. The third data determination unit is used to determine the target sensitive field from the second target data based on the preset field sensitivity conditions, and to perform preset desensitization operation and preset prompt word injection operation on the target sensitive field in the second target data to obtain the corresponding third target data; The context data determination unit is used to generate financial context data based on the third target data and the execution context.
[0052] In some specific embodiments, the fragment buffer module 12 specifically includes: The channel creation unit is used to generate a target request that is adapted to the large language model in the financial business processing agent based on the financial context data, and to create a streaming channel; The fragment generation unit is used to receive streaming financial fragments generated by the large language model based on the target request using the streaming channel.
[0053] In some specific embodiments, the second data generation module 13 specifically includes: An object determination unit is used to determine the skill identifier, parameter information, and target object in the candidate actions, so as to determine the action object of the candidate actions; The first action detection unit is used to determine the target skill corresponding to the skill identifier, and to determine whether there is a skill execution condition corresponding to the target skill on the financial business web terminal, so as to complete the first detection operation of the candidate action; The second action detection unit is used to determine whether the parameter information meets the preset skill declaration structure in order to complete the second detection operation of the candidate action; The third action detection unit is used to determine whether the risk level of the candidate action meets the preset risk range, so as to complete the third detection operation of the candidate action. The data generation unit is used to determine the candidate actions that have passed the detection based on the first detection operation, the second detection operation, and the third detection operation, so as to generate corresponding structured financial data through the skill executor in the financial business web terminal based on the action object and the candidate actions that have passed the detection.
[0054] In some specific embodiments, the second data generation module 13 further includes: The budget determination unit is used to set the current state of the state machine to a preset rollback state if the candidate action detection fails, and to deduct the retry budget to determine the retry budget after deduction. A state setting unit is used to set the current state of the state machine to the preset thinking state if the deducted retry budget meets the preset availability conditions.
[0055] In some specific embodiments, the result generation module 14 specifically includes: The segment rendering unit is used to perform integrity and legality checks on the first target segment in the preset display channel, and to perform preset rendering operations on the first target segment that passes the checks to obtain the corresponding rendering data. The result determination unit is used to determine the financial business processing result corresponding to the financial business request based on the rendered data and the structured financial data.
[0056] Furthermore, embodiments of this application also disclose an electronic device, Figure 6 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the agent-based financial transaction processing method disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.
[0057] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.
[0058] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.
[0059] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the agent-based financial business processing method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.
[0060] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned agent-based financial business processing method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.
[0061] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.
[0062] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0063] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.
[0064] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0065] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. An agent-based financial service processing method, characterized by, A financial business processing intelligent agent applied to the web-based financial business platform, characterized in that it includes: Based on the acquired financial business request, the state machine performs corresponding initialization operations to generate an execution context based on the financial business request and the initialization operations, and uses the financial business request and the execution context to generate financial context data; the state machine is a shared state machine of the financial business processing agent and the financial business web client. Based on the financial context data, a corresponding streaming financial segment is determined, and a first target segment in the streaming financial segment is buffered to a preset display channel, while a second target segment in the streaming financial segment is buffered to a preset inference channel; the first target segment is a financial segment that can be directly displayed on the financial business web terminal; the second target segment is a financial segment that needs to be decided. In the preset reasoning channel, candidate actions are determined based on the second target fragment, and the candidate actions are detected to generate corresponding structured financial data based on the detected candidate actions. Based on the structured financial data and the first target fragment in the preset display channel, a financial business processing result corresponding to the financial business request is generated; The step of performing corresponding initialization operations on the state machine based on the acquired financial business request, to generate an execution context based on the financial business request and the initialization operations, includes: Obtain a financial business request and create a corresponding session context based on the financial business request; initialize the current state of the state machine in the financial business web client to a preset thinking state, and initialize the retry budget and preset log cache corresponding to the financial business request to complete the corresponding initialization operations; generate an execution context based on the financial business request, the session context, and the initialization operations. The step of determining the corresponding streaming financial segment based on the financial context data includes: Based on the financial context data, a target request is generated that is adapted to the large language model in the financial business processing agent, and a streaming channel is created; the streaming channel is used to receive streaming financial segments generated by the large language model based on the target request; The step of detecting the candidate actions to generate corresponding structured financial data based on the detected candidate actions includes: The process involves: identifying the skill identifier, parameter information, and target object in the candidate actions to determine the action object of the candidate actions; identifying the target skill corresponding to the skill identifier and determining whether there are skill execution conditions corresponding to the target skill on the financial business web terminal to complete the first detection operation of the candidate actions; determining whether the parameter information meets a preset skill declaration structure to complete the second detection operation of the candidate actions; determining whether the risk level of the candidate actions meets a preset risk range to complete the third detection operation of the candidate actions; and determining the candidate actions that pass the detection based on the first, second, and third detection operations, so that the skill executor in the financial business web terminal can generate corresponding structured financial data based on the action object and the candidate actions that pass the detection.
2. The financial transaction processing method based on intelligent agents according to claim 1, characterized in that, The step of generating financial context data using the financial transaction request and the execution context includes: Determine the current page's Uniform Resource Locator, routing information, component status, and user selection information to determine the first target data for the financial business web application; The first target data is cropped based on preset data relevance conditions to obtain the second target data; the preset data relevance conditions are the conditions for judging the relevance between the first target data and the financial business request. Based on preset field sensitivity conditions, target sensitive fields are determined from the second target data, and preset desensitization and preset prompt word injection operations are performed on the target sensitive fields in the second target data to obtain the corresponding third target data; Financial context data is generated based on the third target data and the execution context.
3. The financial transaction processing method based on intelligent agents according to claim 1, characterized in that, After detecting the candidate actions, the method further includes: If the candidate action fails the detection, the current state of the state machine is set to the preset rollback state, and the retry budget is deducted to determine the retry budget after deduction. If the deducted retry budget meets the preset availability conditions, then the current state of the state machine is set to the preset thinking state.
4. The agent-based financial transaction processing method according to any one of claims 1 to 3, characterized in that, The step of generating a financial transaction processing result corresponding to the financial transaction request based on the structured financial data and the first target fragment in the preset display channel includes: The first target segment in the preset display channel is subjected to integrity and legality checks, and the first target segment that passes the checks is subjected to preset rendering operations to obtain the corresponding rendering data. The financial transaction processing result corresponding to the financial transaction request is determined based on the rendered data and the structured financial data.
5. A financial transaction processing device based on intelligent agents, characterized in that, Intelligent agents for financial business processing applied to the web-based financial business platform include: The first data generation module is used to perform corresponding initialization operations on the state machine based on the acquired financial business request, so as to generate an execution context based on the financial business request and the initialization operation, and generate financial context data using the financial business request and the execution context; the state machine is a state machine shared by the financial business processing intelligent agent and the financial business web client; The segment buffering module is used to determine the corresponding streaming financial segments based on the financial context data, buffer the first target segment in the streaming financial segment to a preset display channel, and buffer the second target segment in the streaming financial segment to a preset inference channel; the first target segment is a financial segment that can be directly displayed on the financial business web terminal; the second target segment is a financial segment that needs to be decided. The second data generation module is used to determine candidate actions based on the second target fragment in the preset inference channel, and to detect the candidate actions in order to generate corresponding structured financial data based on the detected candidate actions. The result generation module is used to generate a financial business processing result corresponding to the financial business request based on the structured financial data and the first target fragment in the preset display channel; The first data generation module includes: A context creation unit is used to obtain financial business requests and create corresponding session contexts based on the financial business requests; The information initialization unit is used to initialize the current state of the state machine in the financial business web client to a preset thinking state, and to initialize the retry budget and preset log cache corresponding to the financial business request, so as to complete the corresponding initialization operations. A context generation unit is used to generate an execution context based on the financial business request, the session context, and the initialization operation; The fragment buffer module includes: The channel creation unit is used to generate a target request that is adapted to the large language model in the financial business processing agent based on the financial context data, and to create a streaming channel; The segment generation unit is used to receive streaming financial segments generated by the large language model based on the target request using the streaming channel; The second data generation module includes: An object determination unit is used to determine the skill identifier, parameter information, and target object in the candidate actions, so as to determine the action object of the candidate actions; The first action detection unit is used to determine the target skill corresponding to the skill identifier, and to determine whether there is a skill execution condition corresponding to the target skill in the financial business web terminal, so as to complete the first detection operation of the candidate action; The second action detection unit is used to determine whether the parameter information meets the preset skill declaration structure in order to complete the second detection operation of the candidate action; The third action detection unit is used to determine whether the risk level of the candidate action meets the preset risk range, so as to complete the third detection operation of the candidate action. The data generation unit is used to determine the candidate actions that have passed the detection based on the first detection operation, the second detection operation, and the third detection operation, so as to generate corresponding structured financial data based on the action object and the candidate actions that have passed the detection through the skill executor in the financial business web terminal.
6. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the agent-based financial transaction processing method as described in any one of claims 1 to 4.
7. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein, when the computer programs are executed by a processor, they implement the agent-based financial business processing method as described in any one of claims 1 to 4.