Business processing method, apparatus, device, storage medium, and program product

Abstract

This application provides a business processing method applicable to the field of artificial intelligence technology, specifically involving the application of large-scale models in the fintech sector. The method includes: acquiring initial business-related data and performing preprocessing; inputting the preprocessed initial data into a pre-trained business processing model to obtain a target result; and executing the business processing flow based on the target result. The business processing model is a model generated through retrieval enhancement based on a large language model and then reinforced through reinforcement learning. This application also provides a business processing apparatus, device, storage medium, and program product.

Description

Technical Field

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

[0002] In today's financial sector, various financial institutions face both challenges and opportunities related to massive amounts of data. To improve the efficiency and accuracy of business decisions, machine learning models have been widely applied in areas such as credit assessment, fraud prevention, and customer value prediction.

[0003] However, due to the complex and ever-changing business characteristics of the financial sector and the heterogeneity of model construction among different institutions, traditional machine learning methods have significant limitations in model tuning and the transfer and reuse of empirical knowledge. For example, while automated machine learning technology has improved the efficiency of automated modeling to some extent, its search space for hyperparameter combinations is too large, making the parameter tuning process extremely cumbersome. Secondly, existing methods often fail to effectively utilize historical modeling knowledge, making it difficult to effectively transfer and reuse model parameters and parameter tuning approaches across different business lines. These problems not only affect the efficiency of model construction but also limit the efficiency and accuracy of financial institutions' business processing. Summary of the Invention

[0004] In view of the above problems, this application provides a business processing method, apparatus, device, storage medium and program product.

[0005] According to a first aspect of this application, a business processing method is provided, comprising: acquiring initial data related to the business and performing preprocessing; inputting the preprocessed initial data into a pre-trained business processing model to obtain a target result; and executing a business processing flow based on the target result; wherein the business processing model is a model generated by retrieval enhancement based on a large language model and obtained through reinforcement learning.

[0006] According to an embodiment of this application, the business processing model is automatically modeled through reinforcement learning guided by a large language model. The modeling process includes: acquiring the dataset to be trained and performing modular preprocessing; analyzing the features of the dataset to be trained and matching it with a historical case library based on retrieval enhancement generation technology to obtain the target model recommended by the large language model and the corresponding parameter search range.

[0007] According to an embodiment of this application, the modeling process further includes: establishing a two-layer architecture for parameter optimization based on the determination of the target model; wherein the two-layer architecture includes: an architecture layer for Monte Carlo tree-oriented search of discrete parameters; and an architecture layer for applying Bayesian optimization to continuous parameters.

[0008] According to an embodiment of this application, the modeling process further includes: using a near-end policy optimization algorithm based on reinforcement learning, with model performance indicators, computation time, and compliance as multi-objective reward functions, adjusting the modular preprocessing process and the parameters of the target model respectively, to obtain a business processing model.

[0009] According to embodiments of this application, the near-end policy optimization algorithm for reinforcement learning further includes: embedding a Monte Carlo tree search strategy in the action selector.

[0010] According to embodiments of this application, a large language model guides reinforcement learning through a three-dimensional solution space. The construction of the three-dimensional solution space includes: combining operation chains for modular preprocessing in the horizontal dimension; selecting the model architecture in the vertical dimension; and compressing the search space for hyperparameter configuration in the depth dimension.

[0011] According to embodiments of this application, the method further includes compliance verification, which includes: performing compliance checks on all data during the business processing; verifying real-time requirements during the business processing model deployment phase; and performing interpretability verification during the business processing model output phase.

[0012] According to embodiments of this application, the modeling process further includes: constructing a constraint embedding system and a multi-level verification architecture, implementing a real-time correction mechanism and a sandbox verification environment, so as to achieve mathematical completeness in the automatic modeling process of reinforcement learning guided by a large language model.

[0013] According to an embodiment of this application, the business processing flow based on the target result includes: when the evaluation score of the target result is higher than a preset threshold, retaining the current process configuration; when the evaluation score of the target result is lower than the preset threshold, triggering a large language model to guide reinforcement learning to regenerate the business processing model and adjusting the current process configuration.

[0014] The second aspect of this application provides a business processing apparatus, comprising: an initial data acquisition module for acquiring business-related initial data and performing preprocessing; a target result acquisition module for inputting the preprocessed initial data into a pre-trained business processing model to obtain a target result; and a business execution module for executing the business processing flow according to the target result; wherein the business processing model is a model generated by retrieval enhancement based on a large language model and obtained through reinforcement learning.

[0015] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0016] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0017] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method. Attached Figure Description

[0018] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0019] Figure 1 This diagram illustrates an application scenario of a business processing method according to an embodiment of this application.

[0020] Figure 2 A flowchart illustrating a business processing method according to an embodiment of this application is shown schematically.

[0021] Figure 3 A flowchart illustrating the construction of a three-dimensional solution space according to an embodiment of this application is shown schematically.

[0022] Figure 4 A flowchart illustrating the acquisition of a target model according to an embodiment of this application is shown schematically.

[0023] Figure 5 A flowchart illustrating the execution of business processes according to an embodiment of this application is shown schematically;

[0024] Figure 6 This schematically illustrates a structural block diagram of a service processing apparatus according to an embodiment of this application; and

[0025] Figure 7 A block diagram schematically illustrates an electronic device suitable for implementing a business processing method according to an embodiment of this application. Detailed Implementation

[0026] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0027] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0028] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0029] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0030] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0031] In scenarios involving automated decision-making using personal information, the methods, devices, and systems provided in this application all offer users corresponding entry points for choosing to agree to or reject the automated decision-making results. If the user chooses to reject, the process proceeds to the expert decision-making stage. Here, "automated decision-making" refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests, or economic, health, and credit status through computer programs, and then making a decision. Here, "expert decision-making" refers to the activity of making decisions by personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0032] Embodiments of this application provide a business processing method, including: acquiring initial data related to the business and completing preprocessing; inputting the preprocessed initial data into a pre-trained business processing model to obtain a target result; and executing the business processing flow according to the target result; wherein the business processing model is a model generated by retrieval enhancement based on a large language model and obtained through reinforcement learning.

[0033] Through the embodiments of this application, the method optimizes the hyperparameter search space through reinforcement learning guided by a large language model, reduces the time required for model parameter tuning, improves the efficiency of model tuning, and ensures rapid adaptation to different business needs. By combining a large language model and reinforcement learning, the hyperparameter search space is effectively compressed and optimized, significantly reducing the time required for model tuning, improving model construction efficiency, ensuring rapid response to changes in business needs, and improving business processing efficiency and accuracy.

[0034] Figure 1 The diagram illustrates an application scenario of the business processing method according to an embodiment of this application.

[0035] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, and a server 105. Network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0036] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0037] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0038] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0039] It should be noted that the business processing method provided in this application embodiment can generally be executed by server 105. Correspondingly, the business processing device provided in this application embodiment can generally be located in server 105. The business processing method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the business processing device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0040] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0041] The following will be based on Figure 1 The described scene, through Figures 2-5 The business processing method according to the embodiments of this application will be described in detail.

[0042] In the field of banking-related business processing, in order to efficiently and effectively address business needs such as credit assessment, anti-fraud assessment, and value prediction, and to ensure the overall operational efficiency and customer satisfaction of banking operations, this application proposes a business processing method.

[0043] Figure 2 A flowchart illustrating a business processing method according to an embodiment of this application is shown schematically.

[0044] like Figure 2 As shown, according to an embodiment of this application, the business processing method specifically includes operations S210 to S230.

[0045] In operation S210, initial data related to the business is acquired and preprocessed.

[0046] For example, based on the business type, initial data related to that business is retrieved from the information database. For instance, historical transaction data of an account can be retrieved from the transaction record database, including transaction ID, transaction time, transaction amount, and transaction type (deposit, withdrawal, transfer, etc.). Customer credit records, including loan amount, repayment status, and overdue records, can be retrieved from the credit database.

[0047] Furthermore, after obtaining the initial data, data preprocessing is performed. For example, the preprocessing of this initial data includes: first, data cleaning to remove missing values ​​and outliers (such as unreasonable transaction amounts), etc. Then, the data is standardized, for example, by standardizing the format of date fields.

[0048] In operation S220, the preprocessed initial data is input into the pre-trained business processing model to obtain the target result.

[0049] In the embodiments of this application, the business processing model is a model based on retrieval-augmented generation (RAG) of a large language model and pre-trained by reinforcement learning. It is understood that the business processing model is obtained by integrating retrieval-augmented generation (RAG) technology into a large language model and automatically modeling through reinforcement learning guided by the large language model.

[0050] The following section provides a detailed explanation of the construction and reinforcement learning training of this business processing model.

[0051] Figure 3 A flowchart illustrating the construction of a three-dimensional solution space according to an embodiment of this application is shown.

[0052] like Figure 3 As shown in the embodiments of this application, in the process of automatic modeling of reinforcement learning guided by a large language model, the large language model guides reinforcement learning through a three-dimensional solution space. The construction process of the three-dimensional solution space specifically includes operations S310 to S330.

[0053] In operating S310, modular preprocessing operation chain combinations are performed in the horizontal dimension.

[0054] For example, in the process of constructing the three-dimensional solution space, the first step is to perform a modular preprocessing operation chain combination in the horizontal dimension. The preprocessing modules include, for example, a data cleaning module to handle missing values ​​and outliers to ensure data quality; a feature extraction module to extract key features according to business needs, such as customer behavior features and time series features; and a data standardization module to normalize or standardize numerical data to improve the model training effect.

[0055] Furthermore, a large language model is used to generate combination strategies, such as "first cleaning the data, then extracting features, and finally standardizing the data," to design different operation chains and evaluate their impact on subsequent model training. Finally, a reinforcement learning mechanism is used to evaluate the performance of each chain and select the best combination.

[0056] When operating the S320, the model architecture is selected in the vertical dimension.

[0057] For example, first create a library containing various model architectures, such as linear regression, decision trees, random forests, and neural networks. Then, guided by this approach, use a large language model to analyze the current data features, evaluate the applicability of each model architecture, and finally select the most suitable model architecture based on the size of the dataset, the type of features, and the expected task (classification or regression).

[0058] Understandably, reinforcement learning algorithms improve the model's prediction accuracy by continuously trying and receiving feedback to optimize the model architecture selection process.

[0059] When operating the S330, search space compression is performed on hyperparameter configuration in the depth dimension.

[0060] For example, firstly, for the selected model architecture, determine the hyperparameters that need tuning, such as the learning rate, tree depth, and regularization coefficient. Then, analyze historical experimental results using a large language model to identify hyperparameter combinations that significantly impact model performance. Next, use reinforcement learning algorithms to dynamically adjust the initial range of the hyperparameters, reducing unnecessary search time. Finally, advanced methods such as Bayesian optimization can be employed to further focus on potential high-performance hyperparameter regions, improving tuning efficiency.

[0061] Through the embodiments of this application, a highly efficient three-dimensional solution space is constructed by combining the guidance of a large language model and the automation of reinforcement learning. This facilitates modular preprocessing of the model, reasonable model architecture selection, and accurate hyperparameter optimization. This not only improves modeling efficiency but also enhances model performance.

[0062] According to an embodiment of this application, the process of automatically constructing the business processing model using reinforcement learning guided by a large language model will be described in detail below based on the three dimensions of the three-dimensional solution space described above.

[0063] Figure 4 A flowchart illustrating the acquisition of a target model according to an embodiment of this application is shown schematically.

[0064] like Figure 4 As shown, according to an embodiment of this application, based on the three dimensions of the above-mentioned three-dimensional solution space, in the process of automatic modeling through reinforcement learning guided by a large language model, the target model is first acquired. The process of acquiring the target model specifically includes operations S410 to S420.

[0065] The S410 is used to obtain the training dataset and perform modular preprocessing.

[0066] When operating S420, based on the retrieval enhancement generation technology, the features of the dataset to be trained are analyzed and matched with the historical case library to obtain the target model recommended by the large language model and the corresponding parameter search range.

[0067] For example, a training dataset is obtained from a data source. This dataset is understood to be business-related data, such as transaction data or credit data. Data cleaning is then performed, removing invalid records, filling in missing values, and using the aforementioned preprocessing modules for feature extraction. For text data, natural language processing techniques are used for word segmentation and stop word removal; for numerical data, normalization and standardization are performed.

[0068] Furthermore, the large language model is used to analyze the features of the preprocessed dataset, identifying key feature types and relevance. Then, retrieval enhancement generation technology is integrated into the large language model. Through this technology, cases similar to the current dataset are automatically matched from the historical case library, extracting successful modeling experiences. Finally, the large language model recommends target models (such as random forests, deep neural networks, etc.) and their corresponding parameter search ranges (such as learning rate, number of trees, and other hyperparameters).

[0069] In the embodiments of this application, by matching the historical case library, the modeling and parameter tuning experience among the models of various business lines can be effectively reused, reducing redundant work and improving overall efficiency.

[0070] It should be noted that the large language model can be selected from a pre-defined pool of candidate models. This pool of candidate models contains a carefully selected set of various tree models and linear models such as logistic regression.

[0071] In embodiments of this application, the modeling process further includes: establishing a two-layer architecture for parameter optimization based on the determination of the target model; wherein the two-layer architecture includes: an architecture layer for Monte Carlo tree-oriented search of discrete parameters; and an architecture layer for applying Bayesian optimization to continuous parameters.

[0072] Specifically, after selecting the recommended target model, the large language model establishes a two-layer architecture for parameter optimization. In this architecture, the first layer is responsible for optimizing discrete parameters, employing a Monte Carlo tree search strategy to explore and evaluate within the discrete parameter space. The second layer focuses on optimizing continuous parameters, applying Bayesian optimization methods to find the globally optimal hyperparameter combination. Combining these two optimization processes forms an interactive search mechanism.

[0073] Through the embodiments of this application, the two-layer architecture achieves synergistic optimization of the mixed parameter space by using Monte Carlo tree search to optimize discrete parameters and Bayesian optimization to efficiently solve continuous parameters, thereby improving search efficiency.

[0074] In embodiments of this application, the modeling process further includes: using a reinforcement learning-based proximal policy optimization algorithm, with model performance metrics, computation time, and compliance as multi-objective reward functions, adjusting the modular preprocessing flow and the parameters of the target model respectively to obtain a business processing model. Specifically, a Monte Carlo tree search strategy is embedded in the action selector.

[0075] Specifically, a highly efficient business processing model is trained and constructed by combining reinforcement learning's proximal policy optimization algorithm, Monte Carlo tree search, and parameter search techniques. It is understood that in the embodiments of this application, the adaptability and effectiveness of the model are improved through state space modeling, dynamic pruning rules, and the use of a historical scheme caching pool.

[0076] For example, the first step is state space modeling, which integrates the feature-dimensional solution space topology graph of the large model output. By analyzing the features of the training dataset, a multi-dimensional state space is constructed to represent the impact of different feature combinations on model performance. For each state, a feasibility assessment is performed to determine whether the state has reasonable performance potential in practical applications.

[0077] In the embodiments of this application, a near-end policy optimization algorithm is used to dynamically optimize the preprocessing process and model parameters with multiple objectives, including model performance, computational efficiency, and compliance, thereby achieving automated trade-off decisions. This enables an adaptive balance between accuracy and efficiency while ensuring compliance, significantly improving the overall performance and deployment feasibility of the business processing model. Furthermore, the hyperparameter search space is optimized, reducing the time required for model tuning, improving the efficiency of model optimization, and ensuring rapid adaptation to different business needs.

[0078] Furthermore, embodiments of this application also improve Monte Carlo tree search, including: (1) Extending the action selector of the proximal policy optimization algorithm: embedding the Monte Carlo tree search strategy in the action selection part of the proximal policy optimization algorithm. Monte Carlo tree search is used to evaluate the potential benefit of each action, and by simulating future state transitions, the optimal action can be selected more accurately. (2) Setting dynamic pruning rules: when the cumulative reward value is lower than the lower limit value predicted by the large model, the exploration of the branch is automatically terminated to reduce unnecessary computational overhead. This strategy effectively compresses the search space, making the algorithm more efficient. (3) Introducing a historical solution cache pool: loading the top 20% of high-quality solutions as prior knowledge during the initialization phase to form a historical solution cache pool.

[0079] Furthermore, the parameter search technique integrates discrete parameter search and continuous parameter optimization. Discrete parameter search specifically utilizes a large model for pre-screening to generate high-potential preprocessed combinations, thereby reducing ineffective operation chains by 80%. Subsequently, Monte Carlo trees are applied to perform targeted exploration, focusing on nodes with performance gains exceeding 15% to find even better parameter combinations.

[0080] Continuous parameter optimization involves constructing a graph of relationships between parameters (such as the constraint relationship between the learning rate and the regularization coefficient) to help understand the interactions between various hyperparameters. Then, a Bayesian optimizer is used to perform hyperparameter tuning within a constrained space, exploring known local optima in depth to improve the overall model performance.

[0081] Furthermore, during the reinforcement learning process described above, the model is evaluated using a defined multi-objective reward function (including model performance metrics, computation time, and compliance). As training progresses, the strategy is continuously updated based on feedback results, optimizing the modular preprocessing workflow and model parameters to improve the model's adaptability and effectiveness under different settings.

[0082] Finally, after multiple rounds of iterative optimization, the final business processing model was obtained. This model maintains high efficiency and compliance in practical applications and has strong generalization capabilities, adapting to different data environments and task requirements.

[0083] Through the embodiments of this application, a highly efficient business processing model is constructed by combining reinforcement learning's proximal policy optimization algorithm, Monte Carlo tree search, and parameter search techniques. The training efficiency of the model is improved through the application of state space modeling, dynamic pruning, and a historical solution caching pool.

[0084] In the embodiments of this application, the modeling process further includes: constructing a constraint embedding system and a multi-level verification architecture, implementing a real-time correction mechanism and a sandbox verification environment, so as to achieve mathematical completeness in the automatic modeling process of reinforcement learning guided by a large language model.

[0085] Specifically, the first step is to construct a constraint embedding system, mainly including static rule constraints and dynamic boundary constraints. Static rule constraints transform mathematical completeness requirements into a structured rule base (such as preprocessing operation irreversibility constraints and model convergence thresholds), performing hard filtering before the large language model outputs suggestions. Dynamic boundary constraints, based on the current search state of the Monte Carlo tree, generate feasible parameter domains in real time (such as gradient descent step size thresholds and feature combination mutual exclusion rules), dynamically limiting the range of the solution space generated by the large language model through the application programming interface.

[0086] Furthermore, the multi-layered verification architecture mainly includes a syntax verification layer, a semantic verification layer, a mathematical verification layer, and a business verification layer. The syntax verification layer checks whether the proposed solution conforms to basic operational specifications. The semantic verification layer verifies the rationality of the business logic (e.g., anti-fraud models must not use preprocessing methods that reduce recall). The mathematical verification layer runs lightweight mathematical simulations to ensure the theoretical convergence of the optimization process. The business verification layer connects to the banking business rule engine to verify the solution's compliance with regulatory indicators in real time (e.g., feature interpretability score ≥ 0.8).

[0087] Furthermore, a feedback loop is established to collect feedback data in real time during model execution and to check the model using a defined validation architecture. When the output is found to be inconsistent with predetermined constraints or performance metrics, a correction mechanism is activated. This correction mechanism includes adjusting model parameters, reselecting features, or adopting different strategies to bring the model back to an acceptable operating range.

[0088] Furthermore, a sandbox environment is created for offline model validation. In this environment, different input scenarios and constraints are simulated, and the model's response is observed by running it. A large number of test cases are executed in the sandbox environment to ensure the model's stability and reliability. The test results will be used to refine the model and its constraints, ensuring the validated model performs well in real-world environments.

[0089] Finally, after completing the above steps, mathematical completeness evaluation indicators can be designed to conduct a comprehensive analysis of the model's decision-making process.

[0090] By constructing a constraint embedding system, a multi-level verification architecture, and implementing a real-time correction mechanism and a sandbox verification environment, the mathematical completeness of the automatic reinforcement learning modeling process guided by a large language model is effectively guaranteed. Simultaneously, by establishing a multi-layered security verification pipeline, problems such as illegal parameter combinations that may occur in traditional automatic reinforcement learning modeling are avoided.

[0091] When operating S230, the business processing flow is executed based on the target result.

[0092] Figure 5 A flowchart illustrating the execution of business processes according to an embodiment of this application is shown.

[0093] like Figure 5 As shown, according to an embodiment of this application, the process of performing business based on the target result includes operations S510 to S520.

[0094] When operating S510, if the evaluation score of the target result is higher than a preset threshold, the current process configuration is retained.

[0095] When operating S520, if the evaluation score of the target result is lower than the preset threshold, the large language model is triggered to guide reinforcement learning to regenerate the business processing model and adjust the current process configuration.

[0096] Specifically, identify the target outcomes of business processes, such as customer satisfaction and order completion rate. Design corresponding evaluation metrics for these targets. Calculate evaluation scores by analyzing historical data. These scores measure the difference between the model's predicted outcomes and the actual results; higher scores indicate more accurate model predictions.

[0097] After each business processing cycle ends, obtain the evaluation score of the current target result. If the score is higher than the preset threshold (e.g., 0.7), perform the following steps: (1) Record the current process configuration and its corresponding evaluation score to form a backup for subsequent analysis. (2) Keep the current business processing process unchanged and continue to use the existing model for business processing to ensure the continuity and stability of the business. (3) Monitor relevant business indicators and periodically evaluate the effectiveness of the current process configuration.

[0098] Furthermore, if the evaluation score of the target result is lower than a preset threshold, the optimization process of the business processing model needs to be initiated. That is, the reinforcement learning process guided by the large language model is triggered. By learning from historical data and the current business environment, the business processing model is regenerated, and the current process configuration is adjusted. This may include changing input features, optimizing algorithm parameters, or introducing new data sources to improve the model's predictive ability.

[0099] Through the embodiments of this application, a dynamic evaluation and threshold triggering mechanism is used to achieve self-optimization of the business processing flow. High-resolution operations maintain efficient configuration, while low-resolution operations automatically invoke a large language model to guide reinforcement learning for remodeling, forming a closed-loop optimization. This ensures the stability of business processing and enables continuous iterative improvement, significantly enhancing the system's adaptability and long-term performance.

[0100] It should be noted that the process of executing the above business processing methods also includes compliance verification, which includes: conducting compliance checks on all data in the business processing process; verifying real-time requirements during the business processing model deployment phase; and conducting interpretability verification during the business processing model output phase.

[0101] Specifically, during the data collection phase, all input data undergoes preliminary compliance checks. Data quality standards are used to verify the accuracy, completeness, and consistency of the data. For example, a data compliance check module is developed, including the following checks: ensuring the data source is legal and complies with relevant laws and regulations (such as the Personal Information Protection Law); checking whether the data format conforms to predetermined specifications, such as date format and numerical range; and automatically identifying and removing outliers and missing values ​​through data cleaning techniques to ensure the reliability of data in subsequent processing.

[0102] Furthermore, during the deployment phase of the business processing model, real-time requirements are verified to ensure that the model can complete processing tasks within the specified time. Performance benchmarks are established, response time requirements under different business scenarios are defined, and stress testing is performed before model deployment. For example, a simulated environment is created, historical data is used for real-time processing tests, and the model's response time under high load is monitored. If the model fails to meet real-time requirements, the reasons are analyzed, and the algorithm or model structure is optimized to improve processing efficiency.

[0103] Furthermore, after the model generates output, interpretability verification is performed to ensure that users can understand the model's decision-making rationale. For example, an interpretability module can be developed, employing model interpretability techniques to analyze the model's output and provide the basis for decision-making and influencing factors. For instance, for each output, a detailed explanation report can be generated, illustrating which features played a key role in the decision-making process and their specific degree of influence. Visualization tools can also be provided to facilitate users' viewing and understanding of the model's decision-making process.

[0104] Through the embodiments of this application, the method constructs a comprehensive compliance verification system by checking data compliance during business processing, verifying real-time requirements, and verifying the interpretability of model output. This system ensures the legality and compliance of the business processing model at each stage, improving the model's credibility and user satisfaction.

[0105] Based on the above-described business processing method, this application also provides a business processing apparatus. The following will be combined with... Figure 6 The device is described in detail.

[0106] Figure 6 A schematic block diagram of a service processing apparatus according to an embodiment of this application is shown.

[0107] like Figure 6 As shown, the business processing device 600 in this embodiment includes an initial data acquisition module 610, a target result acquisition module 620, and a business execution module 630.

[0108] The initial data acquisition module 610 is used to acquire initial data related to the business and complete preprocessing. In one embodiment, the initial data acquisition module 610 can be used to perform the operation S210 described above, which will not be repeated here.

[0109] The target result acquisition module 620 is used to input the preprocessed initial data into a pre-trained business processing model to obtain the target result. The business processing model is a model generated through retrieval enhancement based on a large language model and then subjected to reinforcement learning. In one embodiment, the target result acquisition module 620 can be used to perform the operation S220 described above, which will not be repeated here.

[0110] The business execution module 630 is used to execute the business processing flow according to the target result. In one embodiment, the business execution module 630 can be used to execute the operation S230 described above, which will not be repeated here.

[0111] In the embodiments of this application, the target result acquisition module 620 includes a business processing model building submodule, which is used to automatically model the business processing model through reinforcement learning guided by a large language model.

[0112] In the embodiments of this application, the business processing model establishment submodule includes a training dataset acquisition unit, a target model acquisition unit, a two-layer architecture establishment unit, a business processing model establishment unit, and a three-dimensional solution space construction unit.

[0113] The training dataset acquisition unit is used to acquire the dataset to be trained and perform modular preprocessing. In one embodiment, the training dataset acquisition unit can be used to perform the operation S410 described above, which will not be repeated here.

[0114] The target model acquisition unit is used to analyze the features of the training dataset and match them with a historical case library based on retrieval enhancement generation techniques to obtain the target model recommended by the large language model and the corresponding parameter search range. In one embodiment, the target model acquisition unit can be used to perform the operation S420 described above, which will not be repeated here.

[0115] A two-layer architecture building unit is used to establish a two-layer architecture for parameter optimization based on the determination of the target model. The two-layer architecture includes: an architecture layer for Monte Carlo tree-oriented search of discrete parameters; and an architecture layer for applying Bayesian optimization to continuous parameters.

[0116] The business processing model building unit is used to optimize the near-end policy through reinforcement learning, using model performance indicators, computation time, and compliance as multi-objective reward functions, to adjust the modular preprocessing process and the parameters of the target model respectively, so as to obtain the business processing model.

[0117] The three-dimensional solution space construction unit is used for combining operation chains for modular preprocessing in the horizontal dimension, selecting the model architecture in the vertical dimension, and compressing the search space for hyperparameter configuration in the depth dimension. In one embodiment, the three-dimensional solution space construction unit can be used to perform operations S310~S330 described above, which will not be repeated here.

[0118] In the embodiments of this application, the service execution module 630 includes a configuration retention submodule and a configuration adjustment submodule.

[0119] The configuration retention submodule is used to retain the current process configuration when the evaluation score of the target result is higher than a preset threshold. In one embodiment, the configuration retention submodule can be used to perform the operation S510 described above, which will not be repeated here.

[0120] The configuration adjustment submodule is used to trigger the large language model-guided reinforcement learning to regenerate the business processing model and adjust the current process configuration when the evaluation score of the target result is lower than a preset threshold. In one embodiment, the configuration adjustment submodule can be used to execute the operation S520 described above, which will not be repeated here.

[0121] In embodiments of this application, the business processing device 600 further includes a compliance verification module. This compliance verification module includes a compliance check submodule, a real-time verification submodule, and an interpretability verification submodule.

[0122] The compliance check submodule is used to perform compliance checks on all data during the business processing.

[0123] The real-time verification submodule is used to verify real-time requirements during the deployment phase of the business processing model.

[0124] The interpretability verification submodule is used to perform interpretability verification during the business processing model output phase.

[0125] In the embodiments of this application, the business processing model is a model generated by retrieval enhancement based on a large language model and obtained through reinforcement learning.

[0126] According to embodiments of this application, any multiple modules among the initial data acquisition module 610, target result acquisition module 620, and service execution module 630 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the initial data acquisition module 610, target result acquisition module 620, and service execution module 630 can be at least partially implemented as hardware circuits, such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits (ASICs), or any other reasonable means of integrating or packaging circuits, or implemented in software, hardware, or firmware, or in any appropriate combination of any of these three implementation methods. Alternatively, at least one of the initial data acquisition module 610, target result acquisition module 620, and service execution module 630 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0127] Figure 7 A block diagram schematically illustrates an electronic device suitable for implementing a business processing method according to an embodiment of this application.

[0128] like Figure 7 As shown, an electronic device 700 according to an embodiment of this application includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage portion 708 into a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0129] RAM 703 stores various programs and data required for the operation of electronic device 700. Processor 701, ROM 702, and RAM 703 are interconnected via bus 704. Processor 701 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 702 and / or RAM 703. It should be noted that programs may also be stored in one or more memories other than ROM 702 and RAM 703. Processor 701 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in one or more memories.

[0130] According to embodiments of this application, the electronic device 700 may further include an input / output (I / O) interface 705, which is also connected to a bus 704. The electronic device 700 may also include one or more of the following components connected to the input / output (I / O) interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the input / output (I / O) interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 710 as needed so that computer programs read from it can be installed into the storage section 708 as needed.

[0131] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0132] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used or combined with an instruction-modified execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 702 and / or RAM 703 and / or one or more memories other than ROM 702 and RAM 703 described above.

[0133] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the business processing methods provided in the embodiments of this application.

[0134] When the computer program is executed by the processor 701, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0135] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 709, and / or installed from a removable medium 711. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0136] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 709, and / or installed from the removable medium 711. When the computer program is executed by the processor 701, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0137] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0138] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0139] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A business processing method, characterized in that, The method includes: Acquire initial business-related data and complete preprocessing; The preprocessed initial data is input into the pre-trained business processing model to obtain the target result; The business processing flow is executed based on the target result; The business processing model is a model generated by retrieval enhancement based on a large language model and obtained through reinforcement learning.

2. The business processing method according to claim 1, characterized in that, The business processing model is automatically obtained through reinforcement learning guided by a large language model. The modeling process includes: Obtain the training dataset and perform modular preprocessing; Based on the retrieval enhancement generation technique, the features of the dataset to be trained are analyzed and matched with the historical case library to obtain the target model recommended by the large language model and the corresponding parameter search range.

3. The business processing method according to claim 2, characterized in that, The modeling process also includes: Based on the determination of the target model, a two-layer architecture for parameter optimization is established; The two-layer architecture includes: an architecture layer for Monte Carlo tree-oriented search of discrete parameters; and an architecture layer for applying Bayesian optimization to continuous parameters.

4. The business processing method according to claim 2 or 3, characterized in that, The modeling process also includes: The business processing model is obtained by using a near-end policy optimization algorithm based on reinforcement learning, with model performance metrics, computation time, and compliance as multi-objective reward functions, and adjusting the modular preprocessing process and the parameters of the target model respectively.

5. The business processing method according to claim 4, characterized in that, The near-end policy optimization algorithm for reinforcement learning further includes: embedding a Monte Carlo tree search strategy in the action selector.

6. The business processing method according to claim 2, characterized in that, The large language model guides reinforcement learning through a three-dimensional solution space. The construction of the three-dimensional solution space includes: In the horizontal dimension, modular preprocessing operation chains are combined. In the vertical dimension, select the model architecture; In the depth dimension, the search space for hyperparameter configuration is compressed.

7. The business processing method according to claim 1, characterized in that, The method further includes compliance verification, which includes: Perform compliance checks on all data during the aforementioned business processing. During the deployment phase of the business processing model, real-time requirements are verified; and Interpretability verification is performed during the output phase of the business processing model.

8. The business processing method according to claim 2, characterized in that, The modeling process also includes: By constructing a constraint embedding system and a multi-level verification architecture, and implementing a real-time correction mechanism and a sandbox verification environment, mathematical completeness is achieved in the automatic modeling process of reinforcement learning guided by the large language model.

9. The business processing method according to claim 1, characterized in that, The process of executing the business based on the target result includes: When the evaluation score of the target result is higher than a preset threshold, the current process configuration is retained; When the evaluation score of the target result is lower than a preset threshold, the large language model is triggered to guide reinforcement learning to regenerate the business processing model and adjust the current process configuration.

10. A business processing apparatus, characterized in that, The device includes: The initial data acquisition module is used to acquire business-related initial data and complete preprocessing. The target result acquisition module is used to input the pre-processed initial data into the pre-trained business processing model to obtain the target result; The business execution module is used to execute the business processing flow according to the target result; The business processing model is a model generated by retrieval enhancement based on a large language model and obtained through reinforcement learning.

11. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 9.

12. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 9.

13. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 9.

Images