Task processing method and related equipment
By dynamically updating and generating task flow templates, the problem of low system scalability in the interconnected banking and enterprise environment is solved, enabling efficient processing to quickly adapt to new business needs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-28
- Publication Date
- 2026-03-10
AI Technical Summary
Existing technologies in the banking and enterprise interconnection environment face problems such as low system scalability, high maintenance costs, and poor flexibility. In particular, when the banking API interface changes, scripts need to be rewritten frequently, making it difficult to quickly adapt to new business needs.
By acquiring template modification information from various banks, the original task flow template is dynamically updated, the target task flow template is generated, and the task flow is dynamically generated and executed based on the target task flow template, thereby achieving dynamic orchestration and standardized processing of task flows.
It improves the system's scalability and flexibility, enabling it to quickly adapt to changes in new business needs, reduces development and maintenance workload, and achieves standardized processing results.
Smart Images
Figure CN121636100A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of this application relate to the field of task processing, and more specifically, to task processing methods, task processing apparatus, task processing devices, computer-readable storage media, and computer program products containing instructions. Background Technology
[0002] In the context of bank-enterprise interconnection (interconnection between banks and enterprises), as transactions and data exchanges between enterprises and banks increase, the complexity and demands of data processing for enterprises also increase. Enterprises need to process more and more bank payment status and interaction data (such as transaction verification, account information, payment instructions, etc.), which requires enterprises to have stronger data processing capabilities.
[0003] The existing task processing method is that when new processing logic changes (such as new business requirements or changes in bank API interfaces), such as when a task for obtaining payment status data is added or modified, developers need to rewrite the scripts for all banks according to the changes in the new processing logic and the data formats of each bank because the field formats of each bank are different. Only after all the scripts for all banks have been rewritten can the rewritten scripts be run to adapt to the changes.
[0004] However, the hard-coding and scripting methods of existing technologies require extensive script rewriting for each change. Furthermore, the rewritten scripts can only be run to adapt to the change after all banks' scripts have been rewritten. This makes it difficult for the system to quickly adapt to new processing logic changes, limiting its scalability. Therefore, existing technologies have low system scalability. Summary of the Invention
[0005] This application provides a task processing method, a task processing apparatus, a task processing device, a computer-readable storage medium, and a computer program product containing instructions for performing task processing while improving system scalability.
[0006] In a first aspect, embodiments of this application provide a task processing method, including:
[0007] Obtain template modification information for each bank, wherein the template modification information for each bank is information on adding, deleting, modifying and / or modifying the dependency order between tasks in the original task flow template of each bank according to the different field formats of each bank.
[0008] Based on the template modification information of each bank, the original task flow templates of each bank are dynamically updated to obtain the target task flow templates of each bank. The target task flow templates of each bank define multiple tasks of each bank and the dependency order between multiple tasks.
[0009] After obtaining a target request to perform specific processing on the target banking business information of the target bank, the target task flow of the target bank is dynamically generated based on the target task flow template of the target bank. The target task flow of the target bank includes multiple target tasks of each target bank and the dependency order between the multiple target tasks.
[0010] By executing the target bank's target task flow to process the target bank's business information, a standardized processing result of the target bank's business information is obtained.
[0011] Secondly, embodiments of this application provide a task processing apparatus, including:
[0012] The acquisition unit is used to acquire template modification information of each bank. The template modification information of each bank includes information on adding, deleting, modifying and / or modifying the dependency order between tasks in the original task flow template of each bank.
[0013] The dynamic update unit is used to dynamically update the original task flow templates of each bank based on the template modification information of each bank, so as to obtain the target task flow templates of each bank. The target task flow templates of each bank define multiple tasks of each bank and the dependency order between multiple tasks.
[0014] The dynamic generation unit is used to obtain a target request for specific processing of the target bank's target banking business information, and then dynamically generate the target task flow of the target bank based on the target task flow template of the target bank. The target task flow of the target bank includes multiple target tasks of each target bank and the dependency order between the multiple target tasks.
[0015] The processing unit is configured to process the target bank's business information by executing the target bank's target task flow, and obtain a standardized processing result of the target bank's business information.
[0016] Thirdly, embodiments of this application provide a task processing device, including:
[0017] Central processing unit, memory, input / output interfaces, wired or wireless network interfaces, and power supply;
[0018] The memory is either a short-term storage memory or a persistent storage memory;
[0019] The central processing unit is configured to communicate with the memory and execute instructions in the memory to perform the aforementioned task processing method.
[0020] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the aforementioned task processing method.
[0021] Fifthly, embodiments of this application provide a computer program product containing instructions that, when run on a computer, cause the computer to perform the aforementioned task processing method.
[0022] As can be seen from the above technical solutions, the embodiments of this application have the following advantages: The original task flow templates of each bank can be dynamically updated based on the template modification information of each bank to obtain the target task flow templates of each bank. After obtaining the target request for specific processing of the target bank's target banking business information, the target task flow of the target bank is dynamically generated based on the target bank's target task flow template. By executing the target bank's target task flow to process the target banking business information, a standardized processing result of the target banking business information is obtained. When the processing logic of new business requirements changes, this dynamic task orchestration method, which dynamically updates task flow templates and dynamically generates task flows, can update and expand the system without affecting the normal operation of existing tasks, and can quickly adapt to new processing logic changes. Therefore, the system of this application has high scalability.
[0023] Accordingly, the task processing apparatus, task processing device, computer-readable storage medium, and computer program product containing instructions provided in this application also have the aforementioned technical effects. Attached Figure Description
[0024] Figure 1 This is a schematic diagram of the architecture of a task processing system disclosed in an embodiment of this application;
[0025] Figure 2-1 This is a flowchart illustrating a task processing method disclosed in an embodiment of this application;
[0026] Figure 2-2 This is a schematic diagram of the overall process of task processing disclosed in an embodiment of this application;
[0027] Figure 2-3 This is a schematic diagram of a task layer architecture disclosed in an embodiment of this application;
[0028] Figure 2-4 This is a schematic diagram of a task layer process disclosed in an embodiment of this application;
[0029] Figure 3 This is a schematic diagram of the structure of a task processing device disclosed in an embodiment of this application;
[0030] Figure 4This is a schematic diagram of the structure of a task processing device disclosed in an embodiment of this application. Detailed Implementation
[0031] This application provides a task processing method, a task processing apparatus, a task processing device, a computer-readable storage medium, and a computer program product containing instructions for performing task processing while improving system scalability.
[0032] The relevant technical features in this field are described as follows:
[0033] DSL: refers to a domain-specific language.
[0034] DAG: Directed Acyclic Graph, is a type of directed graph data structure without loops.
[0035] A reinforcement learning-based agent is an artificial intelligence program that can autonomously discover the best behavioral strategies to achieve its goals by interacting with its environment and learning from rewards or punishments.
[0036] The existing task processing method requires developers to rewrite scripts for all banks when the processing logic for new business requirements changes (e.g., changes to bank API interfaces, such as adding or modifying tasks for obtaining payment status data). Only after all rewritten scripts are completed can the new scripts be run to adapt to the change. This approach has the following drawbacks: First, every change (e.g., changes to bank APIs or business requirements) necessitates rewriting scripts for each bank. This leads to a large amount of repetitive work, and new scripts can only be deployed after all scripts have been rewritten and tested, making it difficult for the system to quickly adapt to new processing logic. Therefore, it suffers from poor scalability and high maintenance costs. Second, the need to maintain independent scripts for each bank, and the potential for modification of multiple scripts with each change, increases maintenance workload and costs. Third, the developed scripts are tightly bound to specific bank interfaces; when the interface changes, the core parts of the script need to be modified, resulting in strong coupling and limiting system flexibility. Fourth, scripts typically only handle problems in specific domains, making them difficult to apply across domains. Modified scripts may affect all related bank-enterprise interactions, resulting in poor flexibility. Fifth, due to the different data structures of different banks, even if the business logic is the same, it is necessary to develop separate scripts for each bank to parse the data. This reduces code reusability, increases the complexity of development and maintenance, and results in poor versatility. Based on this, this application provides a task processing method that dynamically updates the task flow template according to the template modification information of each bank, generates target task flows, and executes these task flows to process banking business information, thereby obtaining standardized processing results. It can be seen that this application can achieve the following effects: First, the system can be updated and expanded without affecting the normal operation of existing tasks, and can quickly adapt to new processing logic changes, resulting in high system scalability. Second, the decoupling of bank and enterprise tasks and dynamic task orchestration enable the system to configure business processing logic, providing flexibility. Third, the integration of a large model engine into the bank and enterprise task layer allows the system to cover both long and short tail requirements and handle cross-domain business needs, providing versatility. Fourth, the layered architecture breaks the "strong binding between scripts and business / interfaces," allowing the system to retain only core data, avoiding coupling. The pluggable nature of bank and enterprise tasks and the dynamically configurable declarative bank and enterprise task flows mean that changes to business logic only require configuration modifications without modifying the core code, providing scalability. Fifth, the bank-enterprise task supervisor can enable continuous optimization of the system.
[0037] Please see Figure 1 The architecture of the task processing system in this application embodiment includes:
[0038] The task processing device 101 and client 102 are configured. When processing tasks, the task processing device 101 can connect to the client 102. The task processing device 101 can obtain template modification information from various banks, dynamically update the original task flow templates of each bank based on this information, and obtain the target task flow templates for each bank. Upon receiving a target request from the client 102 to perform specific processing on target banking business information of the target bank, the task processing device 101 can dynamically generate the target task flow of the target bank based on its target task flow template. By executing the target task flow of the target bank, the target banking business information is processed, and a standardized processing result for the target banking business information is obtained.
[0039] based on Figure 1 Please refer to the task processing system shown. Figure 2-1 , Figure 2-1 This is a flowchart illustrating a task processing method disclosed in an embodiment of this application. The method includes:
[0040] 201. Obtain template modification information for each bank. The template modification information for each bank includes information on the addition, deletion, modification, and / or modification of the dependency order between tasks in the original task flow template of each bank, according to the different field formats of each bank.
[0041] In one alternative implementation, different banks may use different formats for fields representing the same concept, but these fields logically express the same content. For example, Bank A's "amount," Bank B's "amount," and Bank C's "deposit," although their field formats differ, all express the same concept: "transaction amount." The original task flow template represents the task flow templates previously defined for each bank, defining a series of tasks required to process bank data. The dependency sequence represents the order in which tasks are executed, i.e., some tasks must be executed before or after other tasks. Tasks may include, but are not limited to, the following: 1. Checking the payment status of a transaction, such as whether it was successful, failed, or pending. 2. Extracting the payment amount of the transaction, which may involve formatting and validating the amount. 3. Converting the processing results into a unified, standardized format to ensure output consistency. 4. Parsing complex field content, such as extracting key information from text descriptions. 5. Checking whether the payment amount exceeds a preset threshold, such as whether it exceeds 1000 yuan.
[0042] 202. Based on the template modification information of each bank, dynamically update the original task flow template of each bank to obtain the target task flow template of each bank. The target task flow template of each bank defines multiple tasks of each bank and the dependency order between multiple tasks.
[0043] In one alternative implementation, the target task flow template represents the updated task flow template, which defines the updated sequence of tasks and dependencies for each bank.
[0044] 203. After obtaining the target request to perform specific processing on the target bank's target banking business information, dynamically generate the target bank's target task flow based on the target bank's target task flow template. The target bank's target task flow includes multiple target tasks of each target bank and the dependency order between multiple target tasks.
[0045] In one alternative implementation, the target banking business information represents business information (transfer requests, account balance inquiries, payment instructions, etc.) obtained from a specific bank. This information is the object of task flow processing and needs to be identified and processed by the system to complete the business interaction between the bank and the enterprise.
[0046] Specific processing refers to operating on these target banking business information according to specific processing logic (such as new processing logic corresponding to new business requirements or changes in the bank's API interface). For example, if a bank updates its API to include a new check step, "check whether the expected payment time of the payment instruction sent to the bank is consistent with expectations; if not, payment will not be possible," this new check step is specific processing.
[0047] 204. Process the target bank's business information by executing the target bank's target task flow to obtain standardized processing results of the target bank's business information.
[0048] In one alternative implementation, the standardized processing result represents the processing result in a uniform format obtained after executing the task flow, which facilitates further data processing and analysis.
[0049] In this way, the original task flow templates of each bank can be dynamically updated based on the template modification information of each bank, resulting in the target task flow templates of each bank. After obtaining the target request for specific processing of the target bank's target banking business information, the target task flow of the target bank is dynamically generated based on the target bank's target task flow template. By executing the target bank's target task flow, the target banking business information is processed, and a standardized processing result of the target banking business information is obtained. When the processing logic of new business requirements changes, this dynamic task orchestration method, which dynamically updates task flow templates and dynamically generates task flows, can update and expand the system without affecting the normal operation of existing tasks, and can quickly adapt to new changes in processing logic. Therefore, the system of this application has high scalability.
[0050] In one optional implementation, the template modification information for the target bank includes information on adding, deleting, modifying, and / or modifying the dependency order between tasks in the original task flow template of the target bank. Dynamically generating the target task flow of the target bank based on the target task flow template of the target bank includes: dynamically adding, deleting, modifying, and dynamically modifying the dependency order between tasks in the original task flow of the target bank based on the target task flow template of the target bank to generate the target task flow of the target bank. The original task flow of the target bank is generated based on the original task flow template of the target bank.
[0051] This dynamic adjustment mechanism enables the system to respond quickly to changes in business needs without rewriting the entire task flow script, thereby improving the system's flexibility and scalability and reducing development and maintenance workload.
[0052] In one optional implementation, after processing the target bank's business information by executing the target bank's target task flow and obtaining the target bank's standardized processing results, the method further includes: using an intelligent agent to determine the operational status of each target task of the target bank based on the target bank's standardized processing results, and executing reward and punishment actions corresponding to each target task of the target bank based on the operational status of each target task of the target bank, so as to determine the optimization scheme of the target bank's target task flow template, and optimizing the target bank's target task flow template based on the optimization scheme, to obtain the optimized target task flow template of the target bank.
[0053] Specifically, an intelligent agent can be used to evaluate the quality score of each target task in the standardized processing results. The quality score of each target task characterizes its operational status, which includes correctness (whether it meets the expected results), efficiency (task execution speed), and resource consumption (system resources used for task execution). Reward and penalty actions can then be implemented based on the quality scores to optimize the task routing strategy. For example, if the task output is correct and efficient, the agent will provide a positive reward. If the task output is incorrect or inefficient, the agent will provide a negative penalty, thus determining which task to execute at each branch of the task flow, thereby optimizing the overall execution efficiency and quality of the task flow.
[0054] In one example, the agent in this application (a reinforcement learning-based agent) can also be called a bank-enterprise task supervisor. It evaluates the quality of the standardized processing results after the bank-enterprise task is executed, generates a quality score, and implements reward and punishment actions based on the quality score. These actions can directly affect the bank-enterprise task routing strategy, which determines the decision logic of "which task to execute when the task flow generates branches". The system state S observed by the agent is a tuple S = (S1, S2, S3). S1: Execution accuracy of the bank-enterprise task; S2: Characteristics of the current input data (type, complexity, data volume); S3: Downstream feedback (final verification result: correct / incorrect; or manual review label). The agent can execute the following punitive or optimization actions A: A1: Credit downgrade, reducing the task's credit score Credit_Score = Credit_Score * 0.8, directly affecting its probability of being selected for routing. A2: Temporary circuit breaker, setting the task to be disallowed from execution. A3: Trigger retraining, adding the current low-quality output cases to the retraining dataset and asynchronously starting a fine-tuning task. The bank-enterprise task supervisor includes the following reward functions: First, a primary positive reward: "+1.0: The bank-enterprise task output is confirmed correct by the downstream. +0.5: The bank-enterprise task successfully recovers after being penalized (N consecutive successful tasks)." Second, a primary negative reward (penalty): "-1.0: The bank-enterprise task output is confirmed incorrect. -0.7: The bank-enterprise task output has low confidence and ultimately becomes incorrect. -0.3: The task execution times out." The agent learns an optimal strategy by continuously trying S->A->R->S' (updating the system state), that is, learning which action is optimally performed under what system state.
[0055] In this way, after dynamically updating the original task flow template to obtain the target task flow template, the target task flow template can be optimized again based on the standardized processing results, such as optimizing the task execution path, so as to obtain the optimized target task flow template. The final optimized task flow template can better adapt to business needs and improve the accuracy and efficiency of task processing.
[0056] In one optional implementation, after processing the target bank's business information by executing the target bank's target task flow and obtaining the standardized processing result of the target bank's business information, the method further includes: obtaining user feedback information on the standardized processing result, determining an optimization scheme for the target bank's target task flow template based on the feedback information, and optimizing the target bank's target task flow template based on the optimization scheme to obtain the optimized target bank's target task flow template.
[0057] Specifically, feedback information represents users' evaluations or suggestions regarding the system's processing results, including but not limited to feedback on correctness, efficiency, user experience, and improvement suggestions. Optimizing the target bank's target task flow template based on the optimization plan can include, but is not limited to, updating the task flow structure (adjusting the task flow structure based on user feedback, such as adding, deleting, or modifying tasks), adjusting task dependencies (optimizing dependencies between tasks to ensure a more efficient task flow), and updating task logic (improving the task execution logic based on user feedback to enhance task accuracy and efficiency).
[0058] This will better meet user needs and improve the accuracy and efficiency of task processing.
[0059] In one optional implementation, the target task flow of the target bank is dynamically generated based on the target task flow template of the target bank, including: determining the task type of each target task in the target task flow template of the target bank; generating target rule task instances corresponding to each first target task of the first task type; generating target large model task instances corresponding to each second target task of the second task type; and generating the target task flow of the target bank based on the target rule task instances corresponding to each first task and the target large model task instances corresponding to each second task.
[0060] Specifically, the first objective task represents a simple rule-based task, while the second objective task represents a complex text parsing task. By differentiating task types, the system can optimize resource allocation and improve task execution efficiency. Furthermore, the system supports adding new task types and corresponding execution logic without modifying the core code; only adding new task instances is required.
[0061] In one optional implementation, the target bank's target task flow is executed to process the target bank's business information, and a standardized processing result of the target bank's business information is obtained. This includes: executing each target rule task instance using preset rules corresponding to each target rule task instance to obtain a standardized processing result corresponding to each target rule task instance; executing each target large model task instance using the trained target large model corresponding to each target large model task instance to obtain a standardized processing result corresponding to each target large model task instance; and obtaining a standardized processing result of the target bank's business information based on the standardized processing results corresponding to each target rule task instance and each target large model task instance.
[0062] Specifically, the pre-defined rules represent pre-defined task execution logic. These rules are typically formulated based on business needs and data characteristics, and are used to process structured data or perform simple logical judgments. The trained target large model represents the pre-trained and fine-tuned large model, which is capable of handling complex data.
[0063] In this way, pre-defined rule-based task instances provide fast, low-resource-consumption, and high-precision processing capabilities, suitable for handling structured data and simple logical judgments. Large-scale, fully trained model task instances offer powerful generalization and high-precision processing capabilities, suitable for handling complex, unstructured data, such as parsing complex data formats. This design not only improves the efficiency and accuracy of task processing but also enhances the system's flexibility and scalability, enabling it to quickly adapt to new business needs.
[0064] In one optional implementation, before executing each target large model task instance using the trained target large model corresponding to each target large model task instance to obtain the standardized processing result corresponding to each target large model task instance, the method further includes: obtaining banking business information samples from each bank, wherein the banking business information samples are labeled with corresponding specific processing results; using the target large model to determine the predicted specific processing result of the banking business information samples; and obtaining the trained target large model when the loss between the predicted specific processing result and the labeled specific processing result reaches a preset convergence condition.
[0065] Specifically, for large model tasks, domain knowledge can be introduced during execution using the `rag` method to help the large model better understand the background information of a specific domain. Furthermore, the `adapt` method can be used to fine-tune the large model to adapt it to a specific task or dataset. This fine-tuning process requires only a small number of samples to complete the adaptation; for example, only 5 samples are needed: 1 sample to define the input / output structure, 3 samples for fine-tuning training, and 1 sample for validation. In the bank-enterprise task, a prompt word engineering template is configured, automatically obtaining contextual information and injecting it into the template's placeholders, thus forming the final prompt words submitted to the large model. An example of a prompt word template is shown below:
[0066] role: "You are a senior Java ICBC interface expert";
[0067] Task: "Analyze the following code and find the corresponding code for each transaction detail";
[0068] constraints: "Focus only on debit transaction details where dcFlag equals D";
[0069] Examples: [...];
[0070] input_data: ${raw_input} # The system automatically injects this variable.
[0071] In this way, by leveraging the generalization capabilities of large models, and through fine-tuning and prompt word engineering, the system can quickly adapt to new business needs, process complex and unstructured data, and generate standardized processing results. This design not only improves the efficiency and accuracy of task processing but also enhances the system's flexibility and scalability.
[0072] To facilitate understanding of the embodiments of this application, the overall process, task layer architecture, and task layer process of this application are described below.
[0073] Please refer to the details. Figure 2-2 , Figure 2-2 This is a schematic diagram of the overall process of task processing disclosed in an embodiment of this application. Figure 2-2 It can be seen that, Figure 2-2 This paper illustrates the overall workflow design of the task processing system (bank-enterprise interconnection system) in this application, which is divided into five main parts: input layer, bank-enterprise task layer, standardized output layer, bank-enterprise task supervisor, and manual intervention. The input layer is responsible for receiving and preprocessing data from banks. The bank-enterprise task layer is used to execute specific data processing tasks, such as parsing payment status and amount. The standardized output layer is used to convert the processed data into a unified format for subsequent use. The bank-enterprise task supervisor is used to monitor the results of task execution, score and adjust the process to optimize task execution strategies. Manual intervention will be implemented when necessary, such as handling abnormal situations or verifying the supervisor's decisions. For example: Suppose a company needs to process payment data from different banks. The input layer receives data from bank A: {"status": "success", "amount": 1000}. The bank-enterprise task layer parses the data, extracts the payment status and amount, and the standardization output layer formats the data as: {"payment_status": "paid", "amount": 1000}. The bank-enterprise task supervisor evaluates the accuracy of task execution. If errors are found, the task parameters are adjusted or retraining is triggered. Human intervention is provided when the supervisor cannot resolve issues, such as verifying payment statuses described in complex text.
[0074] Please refer to the details. Figure 2-3 , Figure 2-3 This is a schematic diagram of a task layer architecture disclosed in an embodiment of this application, by Figure 2-3 It can be seen that, Figure 2-3This section showcases the architecture of the bank-enterprise task layer, including task templates, a DSL interpreter, a task scheduler, and multiple Large Model Instances (LLMs). Architecture Description: Task templates define a set of reusable task templates for handling different types of data. The bank-enterprise DSL interpreter interprets the task flow templates, converting them into executable tasks. The task scheduler is responsible for scheduling and executing tasks, managing their execution order and dependencies. Large Model Instances (LLMs) handle tasks requiring advanced reasoning or parsing, such as parsing complex text descriptions. Example: A task template defines a template for parsing payment status and amount. The bank-enterprise DSL interpreter reads the template and generates specific task instructions. The task scheduler schedules tasks according to these instructions, such as executing `parse_status` first, then `parse_amount`. Large Model Instances (LLMs) are invoked when needed, such as parsing a complex text description for bank C: `{"description": "Transaction completed successfully"}`. In summary, this architecture supports dynamic task orchestration and execution, as well as handling complex tasks through large models.
[0075] Please refer to the details. Figure 2-4 , Figure 2-4 This is a schematic diagram of a task layer process disclosed in an embodiment of this application, by... Figure 2-4 It can be seen that, Figure 2-4The task layer process design is described in detail, including task orchestration, task execution, and result feedback. Process description: The system initiates the task flow by writing a task flow template. The task orchestration engine can obtain the task flow template based on request data. The task flow template is used to obtain tasks and generate an executable DAG task flow. The task instance repository is used to store and manage task instances. The large model is used to execute large model tasks and return the model execution results. The dynamic task orchestration engine's responsibility is to dynamically generate executable task flows in the bank-enterprise task layer based on the task flow template. It can use a declarative YAML dialect for task flow orchestration as a DSL to write task flow templates. The dynamic task orchestration engine extracts features, identifies metadata from the input data, and automatically matches predefined abstract task flow templates. It instantiates the bank-enterprise task nodes in the template into executable units, automatically infers and establishes dependency edges between tasks based on the dependencies and rules defined in the task flow template, forming an executable DAG. The scheduler then executes the created bank-enterprise task flow. The task flow template can be parsed by the bank-enterprise DSL interpreter. The template contains variable placeholders, and during parsing, it is reconstructed based on contextual information. The template consists of several core blocks: triggers (defining trigger conditions), tasks (defining the content and execution sequence of the bank-enterprise task flow), outputs (standardized output definitions), rules (defining dynamic rules), condition (conditions for task execution), depends_on (depending bank-enterprise tasks), on_failur (elasticity strategy, how to handle failures), and Type (defining the task type, whether it's a rule-based task or a large-scale model task). In summary, this design enables the system to process bank-enterprise tasks efficiently and flexibly, generating standardized processing results, improving task processing efficiency and accuracy, and enhancing the system's flexibility and scalability.
[0076] It is worth mentioning that this application adopts a layered architecture of input layer, bank-enterprise task layer, and standardized output layer, separating "data reception - task processing - format output" into independent modules, breaking the "strong binding between scripts and business / interfaces" in existing technologies. In the task layer, data is flexibly processed through dynamic task orchestration and declarative task flow templates, replacing hard-coded scripts. The standardized output layer uses templated adapter technology to automatically convert task results into a unified and standardized format that meets requirements, effectively solving the problem of different data structures between different banks. Secondly, the bank-enterprise task layer, as the core module of the system, is composed of a dynamic task orchestration engine and multiple large model instance engines. Bank-enterprise tasks are independent plugins that implement a unified adaptation interface, covering functions such as task execution logic, fine-tuning mechanisms, and output format definitions. Bank-enterprise tasks are non-linearly related, and plugins can be freely added and deleted, possessing good scalability. Furthermore, the large model instance engine in the task layer can address the "long-tail demand" within the "long-tail demand" framework. "Long-tail demand" refers to less common scenarios, such as a field returned by a bank's API lacking a parsable schema. In such cases, the generalization ability of the large model can be leveraged. Fine-tuning is first performed using the adapt method of the bank-enterprise task plugin, after which feature content can be extracted from the data lacking a parsable schema. Finally, the bank-enterprise task supervisor scores the tasks based on the quality of their execution results. The scoring influences the task execution strategy, thereby optimizing the task flow template.
[0077] For further details, please refer to Figure 3 One embodiment of the task processing device in this application includes:
[0078] The acquisition unit is used to acquire template modification information of each bank. The template modification information of each bank is information on adding, deleting, modifying and / or modifying the dependency order between tasks in the original task flow template of each bank according to the different field formats of each bank.
[0079] The dynamic update unit is used to dynamically update the original task flow templates of each bank based on the template modification information of each bank, so as to obtain the target task flow templates of each bank. The target task flow templates of each bank define multiple tasks of each bank and the dependency order between multiple tasks.
[0080] The dynamic generation unit is used to obtain a target request for specific processing of the target bank's target banking business information, and then dynamically generate the target task flow of the target bank based on the target task flow template of the target bank. The target task flow of the target bank includes multiple target tasks of each target bank and the dependency order between the multiple target tasks.
[0081] The processing unit is configured to process the target bank's business information by executing the target bank's target task flow, and obtain a standardized processing result of the target bank's business information.
[0082] In one alternative implementation, the dynamic generation unit can be used for:
[0083] Based on the target task flow template of the target bank, the original task flow of the target bank is dynamically modified by adding, deleting, and modifying tasks, as well as dynamically modifying the dependency order between tasks, to generate the target task flow of the target bank. The original task flow of the target bank is generated based on the original task flow template of the target bank. The template modification information of the target bank includes information on adding, deleting, modifying tasks and / or modifying the dependency order between tasks in the original task flow template of the target bank.
[0084] In an optional implementation, the task processing apparatus further includes a determining unit, which can be used to:
[0085] The intelligent agent determines the operational status of each target task of the target bank based on the standardized processing results of the target bank, and executes the corresponding reward and punishment actions for each target task of the target bank based on the operational status of each target task of the target bank, so as to determine the optimization scheme of the target task flow template of the target bank, and optimizes the target task flow template of the target bank based on the optimization scheme, so as to obtain the optimized target task flow template of the target bank.
[0086] In one alternative implementation, the determining unit may be used for:
[0087] Obtain user feedback on the standardized processing results, determine an optimization scheme for the target task flow template of the target bank based on the feedback, and optimize the target task flow template of the target bank based on the optimization scheme to obtain the optimized target task flow template of the target bank.
[0088] In one alternative implementation, the dynamic generation unit can be used for:
[0089] The task type of each target task in the target task flow template of the target bank is determined. For each first target task of the first task type, a target rule task instance corresponding to each first target task is generated. For each second target task of the second task type, a target large model task instance corresponding to each second target task is generated. Based on the target rule task instances corresponding to each first task and the target large model task instances corresponding to each second task, the target task flow of the target bank is generated.
[0090] In one alternative implementation, the dynamic generation unit can be used for:
[0091] Each target rule task instance is executed using the preset rules corresponding to each target rule task instance to obtain the standardized processing result corresponding to each target rule task instance. Each target large model task instance is executed using the trained target large model corresponding to each target large model task instance to obtain the standardized processing result corresponding to each target large model task instance. Based on the standardized processing results corresponding to each target rule task instance and each target large model task instance, the standardized processing result of the target banking business information is obtained.
[0092] In one alternative implementation, the acquisition unit may be used for:
[0093] Obtain banking business information samples from various banks. The banking business information samples are labeled with corresponding specific processing results. Use the target large model to determine the predicted specific processing results of the banking business information samples. When the loss between the predicted specific processing results and the labeled specific processing results reaches a preset convergence condition, the trained target large model is obtained.
[0094] For further details, please refer to Figure 4 One embodiment of the task processing device in this application includes:
[0095] Central processing unit 401, memory 405, input / output interface 404, wired or wireless network interface 403, and power supply 402;
[0096] Memory 405 is either a short-term storage memory or a persistent storage memory;
[0097] The central processing unit 401 is configured to communicate with the memory 405 and execute instructions stored in the memory 405 to perform the aforementioned operations. Figure 2-1 The method in the illustrated embodiment.
[0098] Furthermore, embodiments of this application also provide a computer-readable storage medium, which includes instructions that, when executed on a computer, cause the computer to perform the aforementioned... Figure 2-1 The method in the illustrated embodiment.
[0099] Furthermore, embodiments of this application also provide a computer program product containing instructions, which, when run on a computer, causes the computer to perform the aforementioned... Figure 2-1 The method in the illustrated embodiment.
[0100] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.
[0101] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0102] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, 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 coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.
[0103] 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.
[0104] 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.
[0105] 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 USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
Claims
1. A task processing method characterized by, The method comprises: obtaining template modification information of each bank, the template modification information of each bank being information of adding, deleting, modifying tasks and / or modifying the order of dependence between tasks to the original task flow template of each bank according to the different field formats of each bank; dynamically updating the original task flow template of each bank based on the template modification information of each bank to obtain a target task flow template of each bank, the target task flow template of each bank defining a plurality of tasks and the order of dependence between the plurality of tasks of each bank; after obtaining a target request for performing specific processing on target bank business information, dynamically generating a target task flow of the target bank based on the target task flow template of the target bank, the target task flow of the target bank comprising a plurality of target tasks of the target bank and the order of dependence between the plurality of target tasks; processing the target bank business information by executing the target task flow of the target bank to obtain a standardized processing result of the target bank business information.
2. The method of claim 1, wherein, The template modification information of the target bank comprises information of adding, deleting, modifying tasks and / or modifying the order of dependence between tasks to the original task flow template of the target bank; The dynamically generating a target task flow of the target bank based on the target task flow template of the target bank comprises: dynamically adding, deleting, modifying tasks and dynamically modifying the order of dependence between tasks to the original task flow of the target bank based on the target task flow template of the target bank to generate the target task flow of the target bank, the original task flow of the target bank being generated based on the original task flow template of the target bank.
3. The method of claim 1, wherein, After the processing of the target bank business information by executing the target task flow of the target bank to obtain a standardized processing result of the target bank business information, the method further comprises: determining the running status of each target task of the target bank based on the standardized processing result of the target bank by using an agent, and performing a reward and punishment action corresponding to each target task of the target bank based on the running status of each target task of the target bank to determine an optimization scheme of the target task flow template of the target bank, and optimizing the target task flow template of the target bank based on the optimization scheme to obtain an optimized target task flow template of the target bank.
4. The method of claim 1, wherein, After the processing of the target bank business information by executing the target task flow of the target bank to obtain a standardized processing result of the target bank business information, the method further comprises: obtaining feedback information of a user for the standardized processing result, and determining an optimization scheme of the target task flow template of the target bank based on the feedback information, and optimizing the target task flow template of the target bank based on the optimization scheme to obtain an optimized target task flow template of the target bank.
5. The method of claim 1, wherein, The dynamically generating a target task flow of the target bank based on the target task flow template of the target bank comprises: determining the task type of each target task in the target task flow template of the target bank; For each first target task of the first task type, a target rule task instance corresponding to the each first target task is generated; For each second target task of the second task type, a target large model task instance corresponding to the each second target task is generated; Based on the target rule task instance corresponding to the each first task and the target large model task instance corresponding to the each second task, a target task flow of the target bank is generated.
6. The method of claim 5, wherein, The target bank business information is processed by executing the target task flow of the target bank to obtain a standardized processing result of the target bank business information, which includes: Each target rule task instance is executed by using a preset rule corresponding to the each target rule task instance to obtain a standardized processing result corresponding to the each target rule task instance; Each target large model task instance is executed by using a trained target large model corresponding to the each target large model task instance to obtain a standardized processing result corresponding to the each target large model task instance; Based on the standardized processing result corresponding to the each target rule task instance and the standardized processing result corresponding to the each target large model task instance, the standardized processing result of the target bank business information is obtained.
7. The method of claim 6, wherein, Before the each target large model task instance is executed by using the trained target large model corresponding to the each target large model task instance to obtain the standardized processing result corresponding to the each target large model task instance, the method further includes: Obtaining bank business information samples of each bank, wherein the bank business information samples are labeled with corresponding specific processing results; Using a target large model to determine a predicted specific processing result of the bank business information samples; When the loss between the predicted specific processing result and the labeled specific processing result reaches a preset convergence condition, a trained target large model is obtained.
8. A task processing apparatus characterized by comprising: It includes: An acquisition unit is configured to acquire template modification information of each bank, wherein the template modification information of each bank is information of adding, deleting, modifying tasks and / or modifying the order of task dependencies of an original task flow template of each bank according to different field formats of each bank; A dynamic updating unit is configured to dynamically update the original task flow template of each bank based on the template modification information of each bank to obtain a target task flow template of each bank, wherein the target task flow template of each bank defines a plurality of tasks and the order of task dependencies of each bank; A dynamic generation unit is configured to, after obtaining a target request for performing specific processing on target bank business information of a target bank, dynamically generate a target task flow of the target bank based on the target task flow template of the target bank, wherein the target task flow of the target bank includes a plurality of target tasks of the target bank and the order of target task dependencies; A processing unit is configured to obtain a standardized processing result of the target bank business information by executing the target task flow of the target bank to process the target bank business information.
9. A task processing device characterized by comprising: It includes: A central processing unit and a memory; The memory is a volatile memory or a persistent memory; The central processing unit is configured to communicate with the memory and execute instruction operations in the memory to perform the method of any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer readable storage medium comprises instructions which, when executed on a computer, cause the computer to perform the method of any one of claims 1-7.
11. A computer program product comprising instructions, characterized in that, The computer program product, when executed on a computer, causes the computer to perform the method of any one of claims 1-7.