Financial problem processing method, device and equipment based on double-loop architecture and storage medium

Abstract

The invention discloses a financial problem processing method, device and equipment based on a dual-loop architecture and a storage medium, relates to the technical field of artificial intelligence, is applied to the dual-loop architecture comprising a first loop architecture and a second loop architecture, and comprises the following steps: the first loop architecture receives and analyzes a user financial problem, the second loop architecture generates financial problem processing flow work packages including execution structures and execution steps, calls an evaluation device to perform structure, execution and comparison evaluation on each work package to obtain evaluation results, integrates the evaluation results into feedback information through a feedback loop, and generates a model by using a corpus; processing the evaluation result into a structured training corpus, storing the structured training corpus into a corpus storage library, training the initial large language model by utilizing a low-rank matrix training method and combining the training corpus and feedback information to obtain a superposition model when a condition is met, and fusing the superposition model into a preset reasoning model to obtain a target reasoning model, and the original work package is optimized to generate the new work package, so that the problem processing efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, and storage medium for processing financial problems based on a dual-loop architecture. Background Technology

[0002] Currently, with the development of artificial intelligence technology in recent years, especially breakthroughs in the field of Natural Language Processing (NLP), large language models have demonstrated powerful capabilities in dialogue systems, machine translation, and text generation. Current large language models are primarily used in multi-turn dialogue formats. However, the quality of their responses is highly dependent on user input; overly vague and general question descriptions often lead to uncontrollable and low-quality output. This is particularly true in complex problem domains, where users struggle to quickly and consistently obtain truly high-quality output.

[0003] Therefore, cue word engineering was developed, which involves carefully designing cue statements input to the model to guide it in generating outputs that better meet the requirements. While cue word engineering provides many effective cue word design principles, such as limiting output format, few-sample cues, logical chains, and directional stimuli, it often suffers from the following problems, hindering the full potential of large language models when faced with complex problems:

[0004] Difficult to control precisely: Although prompts can guide the model to produce the desired results, it is difficult to precisely control the specific content of the output, especially for complex or abstract topics.

[0005] Requires expertise: Effective prompt design often requires the knowledge and experience of domain experts, which is a challenge for ordinary users.

[0006] Inefficient: For each new task or scenario, the prompts need to be redesigned and tested, which consumes a lot of time and effort.

[0007] Limited generalization ability: Even if a set of effective prompts is found, they may only be applicable to specific types of input or scenarios and are difficult to apply to a wide range of situations.

[0008] Unforeseen circumstances may arise, such as errors during problem solving, tool failures, or situations where intermediate results are required to determine subsequent execution steps.

[0009] As can be seen from the above, how to improve the efficiency of problem-solving in the process of financial problem handling based on the dual-circulation architecture is an urgent problem to be solved. Summary of the Invention

[0010] In view of this, the purpose of this invention is to provide a method, apparatus, device, and storage medium for processing financial problems based on a dual-loop architecture, which can improve the efficiency of problem processing in the process of processing financial problems based on a dual-loop architecture. The specific solution is as follows:

[0011] Firstly, this application provides a financial problem-solving method based on a dual-loop architecture, applicable to a dual-loop architecture including a first loop architecture and a second loop architecture, comprising:

[0012] The system calls the input interface in the first loop architecture to receive user financial questions, analyzes the user financial questions using a preset inference model, obtains analysis results, and generates several financial question processing workflow packages based on the analysis results, including execution structures and execution steps. The execution structure describes the sequential process of solving the user financial questions, and the execution steps describe the specific operation content and operation results of each process node.

[0013] The first loop architecture and the preset evaluation device are invoked to perform structural evaluation, execution evaluation and comparative evaluation on each of the financial problem processing workflow packages, respectively, to obtain target evaluation results including the first evaluation result, the second evaluation result and the third evaluation result, and the target evaluation results are integrated using the preset feedback loop to obtain feedback information;

[0014] The second loop architecture and the preset corpus generation model are invoked, and the target evaluation results are processed based on the preset format to obtain structured training corpus, which is then stored in the preset corpus repository.

[0015] The second loop architecture is invoked to determine whether each training corpus in the preset corpus repository meets the preset conditions. If it does, the low-rank matrix training method is used to train the initial large language model based on the training corpus and the feedback information to obtain the superimposed model. The superimposed model is then fused into the preset inference model to obtain the target inference model. The target inference model is then used to adjust the financial problem processing workflow package to obtain a new financial problem processing workflow package.

[0016] Optionally, the step of calling the input interface in the first loop architecture to receive user financial questions, analyzing the user financial questions using a preset inference model, obtaining analysis results, and generating several financial question processing workflow packages including execution structures and execution steps based on the analysis results, including:

[0017] The system uses the input interface in the first loop architecture to receive user financial questions and calls the problem analysis unit in the preset reasoning model to analyze the user financial questions, and obtains analysis results including the essence of the problem and the problem requirements.

[0018] The structure generation unit connected to the problem analysis unit is invoked to generate an execution structure corresponding to the user's financial problem based on the analysis results.

[0019] The step filling unit connected to the structure generation unit is called and specific operation instructions and execution results corresponding to each step are generated sequentially based on the execution order corresponding to the execution structure to obtain the execution steps, so as to generate a financial problem processing workflow package based on the execution structure and the execution steps.

[0020] Determine the defect information corresponding to the financial problem handling process work package, and call the iterative improvement unit and feedback loop to adjust, correct and reconstruct the execution structure and execution steps in the financial problem handling process work package generated in the previous round based on the defect information, so as to generate a new financial problem handling process work package.

[0021] Optionally, the first loop architecture and the preset evaluation device are invoked to perform structural evaluation, execution evaluation, and comparative evaluation on each of the financial problem processing workflow packages, respectively, to obtain target evaluation results including a first evaluation result, a second evaluation result, and a third evaluation result. The target evaluation results are then integrated using a preset feedback loop to obtain feedback information, including:

[0022] The structural evaluation model is invoked, and the feasibility, coherence, and correctness of the execution structure are judged using preset logical consistency, preset step completeness, and preset operability rules, resulting in a first evaluation result including a first score and structural defect information; the structural defect information is used to describe the unreasonable information of the execution structure.

[0023] The comparative evaluation model is invoked to receive the financial problem handling process work packages in the same iteration round, and the final completion status, result accuracy, process complexity and expression clarity of each financial problem handling process work package are compared to obtain a second evaluation result including a second score and overall defect information; the second score is a relative merit score corresponding to each financial problem handling process work package.

[0024] The execution evaluation model is invoked to determine the execution results of the execution steps in each of the financial problem handling process work packages and the connection relationship between the steps. Based on the execution results and the connection relationship, a third evaluation result including a third score and alternative solution information is generated. Then, the target evaluation result is integrated using a preset feedback loop to obtain feedback information. The preset evaluation equipment includes a structural evaluation model, a comparative evaluation model and an execution evaluation model.

[0025] Optionally, the step of generating a third evaluation result based on the execution result and the connection relationship, including a third score and alternative solution information, includes:

[0026] Determine whether the execution steps corresponding to each of the financial problem handling process packages need to be distinguished by calling external tools, and obtain the first financial problem handling process package that needs to be distinguished by calling external tools and the second financial problem handling process package that does not need to be distinguished by calling external tools;

[0027] Determine the execution steps and execution records corresponding to each of the first financial problem handling process work packages, and then use preset evaluation criteria to conduct a correctness assessment based on the execution steps and execution records to obtain the corresponding first pending evaluation result;

[0028] Determine the calling instructions and the return results of the corresponding external tools for each of the second financial problem processing workflow packages, and evaluate whether the calling instructions and the return results report errors, meet expectations, and have reasonable execution times to obtain a second pending evaluation result, and generate a third score based on the first pending evaluation result and the second pending evaluation result.

[0029] Determine whether the third score is greater than a preset score threshold. If the third score is greater than the preset score threshold, generate corresponding alternative solution information based on the third score.

[0030] Optionally, the step of calling the second loop architecture and the preset corpus generation model, processing the target evaluation result based on a preset format to obtain structured training corpus, and storing the training corpus in a preset corpus repository includes:

[0031] The second loop architecture is invoked to format the target evaluation result to obtain a formatted evaluation result. Then, a preset corpus generation model is used to extract, reorganize and standardize the information in the formatted evaluation result based on the training specifications corresponding to the training task, thereby generating a structured training corpus.

[0032] Determine the data size corresponding to each training corpus in the preset corpus repository, and determine whether the data size is greater than a preset size threshold. If the data size is not greater than the preset size threshold, then use a preset text file system to store the training corpus in the preset corpus repository.

[0033] If the data size is not greater than the preset size threshold, the preset database system is used to monitor the accumulation status and time status of each training corpus in the preset corpus repository, and the model training process is triggered when the accumulation status meets the first preset condition or the time status meets the second preset condition.

[0034] Optionally, the second loop architecture is invoked to determine whether each training corpus in the preset corpus repository meets a preset condition. If it does, a low-rank matrix training method is used, and the initial large language model is trained based on the training corpus and the feedback information to obtain a superimposed model. The superimposed model is then fused into the preset inference model to obtain a target inference model. This target inference model is then used to adjust the financial problem processing workflow package to obtain a new financial problem processing workflow package, including:

[0035] The second loop architecture is invoked to determine whether each training corpus in the preset corpus repository is greater than a preset corpus quantity threshold, and the determination result is obtained;

[0036] If the judgment result indicates that each of the training corpora is greater than the preset corpus quantity threshold, then a number of training corpora are extracted from the preset corpus repository to obtain training samples. The initial large language model is trained and its parameters are fine-tuned using low-rank matrix factorization training technology and based on the training samples to obtain a superimposed model with incremental knowledge.

[0037] The superimposed model is loaded and fused into the parameter space of the preset inference model used to analyze and generate the financial problem handling process work package to obtain the target inference model. The financial problem handling process work package is then adjusted using the target inference model to obtain a new financial problem handling process work package.

[0038] Secondly, this application provides a financial problem-solving apparatus based on a dual-loop architecture, applicable to a dual-loop architecture including a first loop architecture and a second loop architecture, comprising:

[0039] The work package generation module is used to call the input interface in the first loop architecture to receive user financial questions, analyze the user financial questions using a preset inference model, obtain analysis results, and generate several financial question processing workflow work packages based on the analysis results, including execution structures and execution steps; the execution structure is used to describe the sequential process of solving the user financial questions; the execution steps are used to describe the specific operation content and operation results of each process node.

[0040] The feedback information generation module is used to call the first loop architecture and the preset evaluation device to perform structural evaluation, execution evaluation and comparative evaluation on each of the financial problem processing workflow packages, to obtain target evaluation results including the first evaluation result, the second evaluation result and the third evaluation result, and to integrate the target evaluation results using the preset feedback loop to obtain feedback information.

[0041] The training corpus generation module is used to call the second loop architecture and the preset corpus generation model and process the target evaluation results based on the preset format to obtain structured training corpus, and store the training corpus in the preset corpus repository.

[0042] The model training module is used to call the second loop architecture to determine whether each training corpus in the preset corpus repository meets the preset conditions. If it does, the initial large language model is trained using the low-rank matrix training method based on the training corpus and the feedback information to obtain the superimposed model. The superimposed model is then fused into the preset inference model to obtain the target inference model. The target inference model is then used to adjust the financial problem processing workflow package to obtain a new financial problem processing workflow package.

[0043] Optionally, the feedback information generation module includes:

[0044] The first evaluation result generation unit is used to call the structural evaluation model and use preset logical consistency, preset step completeness and preset operability rules to judge the feasibility, coherence and correctness of the execution structure, and obtain a first evaluation result including a first score and structural defect information; the structural defect information is used to describe the unreasonable information of the execution structure.

[0045] The second evaluation result generation unit is used to call the comparative evaluation model to receive each of the financial problem handling process work packages in the same iteration round, and compare the final completion status, result accuracy, process complexity and expression clarity of each of the financial problem handling process work packages to obtain a second evaluation result including a second score and overall defect information; the second score is a relative merit score corresponding to each of the financial problem handling process work packages.

[0046] The third evaluation result generation unit is used to call the execution evaluation model to determine the execution results of the execution steps in each of the financial problem handling process work packages and the connection relationship between the steps, so as to generate a third evaluation result including a third score and alternative solution information based on the execution results and the connection relationship. Then, the target evaluation result is integrated using a preset feedback loop to obtain feedback information. The preset evaluation equipment includes a structural evaluation model, a comparative evaluation model and an execution evaluation model.

[0047] Thirdly, this application provides an electronic device, comprising:

[0048] Memory, used to store computer programs;

[0049] A processor is used to execute the computer program to implement the aforementioned financial problem-solving method based on a dual-loop architecture.

[0050] Fourthly, this application provides a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned financial problem processing method based on a dual-loop architecture.

[0051] As can be seen from the above, before processing financial problems based on a dual-loop architecture, this application needs to call the input interface in the first loop architecture to receive user financial problems, analyze the user financial problems using a preset inference model, obtain analysis results, and generate several financial problem processing workflow packages including execution structures and execution steps based on the analysis results. Secondly, the first loop architecture and a preset evaluation device are used to perform structural evaluation, execution evaluation, and comparative evaluation on each financial problem processing workflow package, obtaining target evaluation results including a first evaluation result, a second evaluation result, and a third evaluation result. Finally, a preset feedback loop is used to integrate the target evaluation results to obtain feedback. Information; then, the second loop architecture and the preset corpus generation model are invoked, and the target evaluation results are processed based on the preset format to obtain structured training corpus, which is then stored in the preset corpus repository; finally, the second loop architecture is invoked to determine whether each training corpus in the preset corpus repository meets the preset conditions. If it does, the low-rank matrix training method is used, and the initial large language model is trained based on the training corpus and feedback information to obtain the superimposed model. The superimposed model is then integrated into the preset inference model to obtain the target inference model. The target inference model is then used to adjust the financial problem processing workflow package to obtain a new financial problem processing workflow package.

[0052] Therefore, this application first needs to call the input interface in the first loop architecture to receive user financial questions, and then use a preset inference model to analyze the user financial questions, obtain analysis results, and generate several financial question processing workflow packages including execution structures and execution steps based on the analysis results. Second, the first loop architecture and preset evaluation equipment are called to perform structural evaluation, execution evaluation, and comparative evaluation on each financial question processing workflow package to obtain target evaluation results, and the target evaluation results are integrated using a preset feedback loop to obtain feedback information. Then, the second loop architecture and preset corpus generation model are called to process the target evaluation results based on a preset format to obtain structured training corpus, and the training corpus is stored in a preset corpus repository. Finally, the second loop architecture is called to determine whether each training corpus in the preset corpus repository meets the preset conditions. If it does, the low-rank matrix training method is used to train the initial large language model based on the training corpus and feedback information to obtain a superimposed model, and the superimposed model is merged into the preset inference model to obtain the target inference model. The target inference model is then used to adjust the financial question processing workflow packages to obtain new financial question processing workflow packages. This improves the efficiency of problem-solving in the dual-loop architecture-based financial problem-solving process, thereby enhancing the user experience. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0054] Figure 1 This is a flowchart of a financial problem-solving method based on a dual-loop architecture disclosed in this application;

[0055] Figure 2 This is a schematic diagram of a specific dual-loop architecture disclosed in this application;

[0056] Figure 3 This is a schematic diagram illustrating a specific execution structure and execution steps disclosed in this application;

[0057] Figure 4 This is a schematic diagram of a financial problem processing device based on a dual-loop architecture disclosed in this application;

[0058] Figure 5 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Currently, with the development of artificial intelligence technology in recent years, especially breakthroughs in the field of natural language processing, large language models have demonstrated powerful capabilities in dialogue systems, machine translation, and text generation. Current large language models are primarily used in multi-turn dialogue formats. However, the quality of responses from large language models is highly dependent on user input; overly vague and general question descriptions often lead to uncontrollable and low-quality output. Especially in complex problem domains, users find it difficult to quickly and consistently obtain truly high-quality output. Therefore, this application provides a financial problem-solving method based on a dual-loop architecture, which improves the efficiency of problem processing in financial problem-solving within a dual-loop architecture.

[0061] See Figure 1 As shown, this embodiment of the invention discloses a financial problem-solving method based on a dual-loop architecture, applied to a dual-loop architecture including a first loop architecture and a second loop architecture, comprising:

[0062] Step S11: Call the input interface in the first loop architecture to receive the user's financial question, and use a preset inference model to analyze the user's financial question to obtain the analysis result. Based on the analysis result, generate several financial question processing workflow packages including execution structures and execution steps. The execution structure is used to describe the sequential process of solving the user's financial question. The execution steps are used to describe the specific operation content and operation result of each process node.

[0063] In this embodiment, to ensure its effectiveness in both single-task implementations and continuous model improvement, a dual-loop architecture is designed, as illustrated in the diagram below. Figure 2 As shown in the diagram, the smaller loop consists of: inference model -> working package -> evaluation model, which combines a competitive mechanism and immediate evaluation feedback to optimize the output of a single complex task. The larger loop, in turn, consists of: inference model -> working package -> evaluation model -> corpus generation model -> corpus. Its function is to process the feedback results from the smaller loops into training corpora, and to use fine-tuning techniques to train a low-rank superposition model, thereby improving the performance of the inference model. It is worth noting that the key modules in the dual-loop architecture include the inference model, working package, evaluation model, corpus generation model, and corpus.

[0064] It is worth mentioning that the inference model in this embodiment is a superposition model obtained by combining a conventional large language model with a superimposed model trained by fine-tuning training techniques. Its function is to analyze complex questions input by the user and output working packages. In a certain iteration, based on feedback from the evaluation model, the working package from the previous round is improved, and a new working package is output.

[0065] Furthermore, to address complex problems, this application embodiment simplifies the problem by breaking it down into steps. Therefore, this application embodiment restricts the output of the inference model, requiring it to output in a work package format as designed. Each work package includes an execution structure and execution steps. The execution structure describes the specific steps required to solve the problem, while the execution steps provide the specific execution result for each step. Because subsequent steps may depend on the results of previous steps, the generation of execution steps may need to be completed in multiple sessions. The work package is considered complete only when all execution steps have been executed or execution has been interrupted. An example diagram illustrating the execution structure and execution steps is shown below. Figure 3 As shown, in each iteration, not just one work package is generated, but multiple work packages are generated and sent to the evaluation model for evaluation.

[0066] Specifically, the system calls the input interface in the first loop architecture to receive user financial questions, analyzes the questions using a preset inference model, obtains analysis results, and generates several financial problem processing workflow packages based on the analysis results. These packages may include: receiving user financial questions via the input interface in the first loop architecture, analyzing the questions using the problem analysis unit in the preset inference model, obtaining analysis results including the problem's essence and requirements; calling the structure generation unit connected to the problem analysis unit to generate an execution structure corresponding to the user's financial question based on the analysis results; calling the step filling unit connected to the structure generation unit to generate specific operation instructions and execution results corresponding to each step based on the execution order of the execution structure, obtaining execution steps, and generating financial problem processing workflow packages based on the execution structure and execution steps; identifying defect information corresponding to the financial problem processing workflow packages, and calling the iterative improvement unit and feedback loop to adjust, correct, and reconstruct the execution structure and execution steps in the financial problem processing workflow packages generated in the previous round based on the defect information, generating new financial problem processing workflow packages.

[0067] Step S12: Call the first loop architecture and the preset evaluation device to perform structural evaluation, execution evaluation and comparative evaluation on each of the financial problem processing workflow packages, and obtain target evaluation results including the first evaluation result, the second evaluation result and the third evaluation result. Then, use the preset feedback loop to integrate the target evaluation results to obtain feedback information.

[0068] In this embodiment, the evaluation model is the core module of the dual-loop architecture in this application embodiment. It includes three sub-modules: a structural evaluation module, a comparative evaluation module, and an execution evaluation module. In each iteration, the evaluation model receives multiple working packages for that round and uses them to evaluate their effectiveness and provide improvement suggestions. In the smaller loop, its evaluation conclusions are directly fed into the inference model as the basis for generating working packages in the next iteration. In the larger loop, its evaluation conclusions are used as input to the corpus generation model to generate training corpus. The three sub-models are described below:

[0069] Structural Evaluation Model: The structural evaluation model is used to assess whether the step division within each work package is reasonable. The input is the execution structure of each work package; the evaluation criteria are whether the steps within the execution structure are feasible, coherent, and correct; the output is the score and defects of the execution structure.

[0070] Execution evaluation model: used to check the execution results of each execution step, as well as the correctness of the connection between steps.

[0071] Comparative Evaluation Model: The purpose of the comparative evaluation model is to provide an overall assessment by comparing the various work packages generated in this round. The input consists of all work packages generated in this round, and the evaluation criteria are whether the work package was successfully executed, whether the final result is correct, the simplicity of the steps, and the understandability. The output consists of the score and defects of each work package.

[0072] Specifically, the first-loop architecture and preset evaluation equipment are invoked to perform structural evaluation, execution evaluation, and comparative evaluation on each financial problem processing workflow package, respectively, to obtain target evaluation results including a first evaluation result, a second evaluation result, and a third evaluation result. A preset feedback loop is then used to integrate the target evaluation results to obtain feedback information, which may include: invoking the structural evaluation model and using preset logical consistency, preset step completeness, and preset operability rules to judge the feasibility, coherence, and correctness of the execution structure, obtaining a first evaluation result including a first score and structural defect information; the structural defect information is used to describe the unreasonable information of the execution structure; and invoking the comparative evaluation model to receive data from each financial problem processing workflow package in the same iteration round. The problem-solving process work packages are compared in terms of final completion status, result accuracy, process complexity, and clarity of expression to obtain a second evaluation result, including a second score and overall defect information. The second score is a relative merit score corresponding to each financial problem-solving process work package. The execution evaluation model is invoked to determine the execution results of the execution steps in each financial problem-solving process work package and the connection relationship between the steps. Based on the execution results and connection relationship, a third evaluation result including a third score and alternative solution information is generated. Then, the target evaluation results are integrated using a preset feedback loop to obtain feedback information. The preset evaluation equipment includes a structural evaluation model, a comparative evaluation model, and an execution evaluation model.

[0073] It is worth mentioning that the execution evaluation model is used to check the execution results of each execution step and the correctness of the connection between steps. Because the execution steps need to be distinguished according to whether external tools are called: (1) For steps that do not require external tools, the input is the execution step and the execution record, the evaluation criteria are the correctness of the execution of the step and whether the execution result can be used as the input of the next step, and the output is the score of the step. (2) For steps that require external tools, the input is the calling instruction of the step and the return result of the tool, the evaluation criteria are whether the return result reports an error, whether it meets the expectations, and whether the execution time is reasonable. The output is the score of the step, and if there is an error, an alternative solution needs to be included.

[0074] Specifically, a third evaluation result, including a third score and alternative solution information, is generated based on the execution results and the connection relationships. This can include: determining whether the execution steps corresponding to each financial problem handling process package need to be distinguished by calling external tools, thus obtaining a first financial problem handling process package that needs to be distinguished by calling external tools and a second financial problem handling process package that does not need to be distinguished by calling external tools; determining the execution steps and execution records corresponding to each first financial problem handling process package, and then using preset evaluation criteria and based on the execution steps and execution records to evaluate the correctness, thus obtaining the corresponding first pending evaluation result; determining the calling instructions and the return results of the corresponding external tools corresponding to each second financial problem handling process package, and evaluating whether the calling instructions and return results report errors, meet expectations, and have reasonable execution time, thus obtaining a second pending evaluation result, and generating a third score based on the first and second pending evaluation results; determining whether the third score is greater than a preset scoring threshold, and if the third score is greater than the preset scoring threshold, generating corresponding alternative solution information based on the third score.

[0075] Step S13: Invoke the second loop architecture and the preset corpus generation model, process the target evaluation result based on the preset format to obtain structured training corpus, and store the training corpus in the preset corpus repository.

[0076] In this embodiment, the corpus generation model receives the evaluation package from the evaluation model, and then restricts the output format according to the needs of the training corpus to output structured training corpus. Subsequently, the generated training corpus is stored in a corpus. For small corpora with fewer than 30,000 entries, text files can be used for management. For medium to large corpora with more than 30,000 entries, a database system can be used for management. When the corpus accumulates to a certain amount, or after a certain period of time since the last training, the corpus is input into the superposition model training program for training.

[0077] Specifically, the process of calling the second loop architecture and the preset corpus generation model, processing the target evaluation results based on a preset format to obtain structured training corpus, and storing the training corpus in a preset corpus repository may include: calling the second loop architecture to format the target evaluation results to obtain formatted evaluation results; then using the preset corpus generation model and based on the training specifications corresponding to the training task to extract, reorganize, and standardize the information in the formatted evaluation results to generate structured training corpus; determining the data size corresponding to each training corpus in the preset corpus repository and judging whether the data size is greater than a preset size threshold; if the data size is not greater than the preset size threshold, storing the training corpus in the preset corpus repository using a preset text file system; if the data size is not greater than the preset size threshold, monitoring the accumulation status and time status of each training corpus in the preset corpus repository using a preset database system, and triggering the model training process when the accumulation status meets a first set condition or the time status meets a second set condition.

[0078] Step S14: Call the second loop architecture to determine whether each training corpus in the preset corpus repository meets the preset conditions. If it does, use the low-rank matrix training method and the training corpus and the feedback information to train the initial large language model to obtain the superimposed model. Then, merge the superimposed model into the preset inference model to obtain the target inference model. Use the target inference model to adjust the financial problem processing workflow package to obtain a new financial problem processing workflow package.

[0079] In this embodiment, the workflow of the small loop is as follows: First, the user inputs a question, and the inference model provides a fixed number of work packages based on the question. The execution structure and specific execution steps of these work packages are input into the evaluation model for evaluation. The three sub-models of the evaluation model provide evaluation results, which are then integrated to provide feedback to the inference model. The next iteration continues until the result meets expectations. Furthermore, the initial workflow of the large loop is consistent with that of the small loop. However, there is a difference in the processing of the evaluation model's output. The large loop inputs the evaluation results into the corpus generation model, which processes them to form structured training corpus, which is then stored in the corpus. When the training corpus in the corpus reaches a certain quantity, training is initiated using a mature low-rank matrix training method to train an overlay model, which is then overlaid on the inference model. The system then waits for the next round of training to update the overlay model.

[0080] Specifically, the step of calling the second loop architecture to determine whether each training corpus in the preset corpus repository meets a preset condition. If it does, the initial large language model is trained using a low-rank matrix training method based on the training corpus and the feedback information to obtain a superimposed model. The superimposed model is then fused into the preset inference model to obtain a target inference model. This target inference model is then used to adjust the financial problem processing workflow package to obtain a new financial problem processing workflow package. This step may include: calling the second loop architecture to determine whether each training corpus in the preset corpus repository exceeds a preset corpus quantity threshold. A judgment result is obtained; if the judgment result indicates that each training corpus is greater than the preset corpus quantity threshold, then several training corpora are extracted from the preset corpus repository to obtain training samples, and the initial large language model is trained and its parameters are fine-tuned using low-rank matrix factorization training technology and based on the training samples to obtain an overlay model with incremental knowledge; the overlay model is loaded and fused into the parameter space of the preset inference model used to analyze and generate the financial problem processing workflow package to obtain the target inference model, so as to adjust the financial problem processing workflow package using the target inference model to obtain a new financial problem processing workflow package.

[0081] As can be seen from the above, the embodiments of this application first need to call the input interface in the first loop architecture to receive user financial questions, and use a preset inference model to analyze the user financial questions, obtain analysis results, and generate several financial question processing workflow packages including execution structures and execution steps based on the analysis results; secondly, call the first loop architecture and the preset evaluation device to perform structural evaluation, execution evaluation and comparative evaluation on each financial question processing workflow package to obtain target evaluation results, and use a preset feedback loop to integrate the target evaluation results to obtain feedback information; then, call the second loop architecture and the preset corpus generation model and process the target evaluation results based on a preset format to obtain structured training corpus, and store the training corpus in a preset corpus repository; finally, call the second loop architecture to determine whether each training corpus in the preset corpus repository meets the preset conditions. If it does, use the low-rank matrix training method and use the training corpus and feedback information to train the initial large language model to obtain a superimposed model, and then merge the superimposed model into the preset inference model to obtain the target inference model, so as to adjust the financial question processing workflow packages using the target inference model to obtain new financial question processing workflow packages. This improves the efficiency of problem-solving in the dual-loop architecture-based financial problem-solving process, thereby enhancing the user experience.

[0082] Accordingly, see Figure 4As shown, this application also provides a financial problem-solving apparatus based on a dual-loop architecture, applied to a dual-loop architecture including a first loop architecture and a second loop architecture, comprising:

[0083] The work package generation module 11 is used to call the input interface in the first loop architecture to receive user financial questions, analyze the user financial questions using a preset inference model, obtain analysis results, and generate several financial question processing workflow work packages based on the analysis results, including execution structures and execution steps; the execution structure is used to describe the sequential process of solving the user financial questions; the execution steps are used to describe the specific operation content and operation results of each process node.

[0084] The feedback information generation module 12 is used to call the first loop architecture and the preset evaluation device to perform structural evaluation, execution evaluation and comparative evaluation on each of the financial problem processing workflow packages, to obtain target evaluation results including the first evaluation result, the second evaluation result and the third evaluation result, and to integrate the target evaluation results using the preset feedback loop to obtain feedback information.

[0085] The training corpus generation module 13 is used to call the second loop architecture and the preset corpus generation model and process the target evaluation result based on the preset format to obtain structured training corpus, and store the training corpus in the preset corpus repository.

[0086] The model training module 14 is used to call the second loop architecture to determine whether each training corpus in the preset corpus repository meets the preset conditions. If it does, the initial large language model is trained using the low-rank matrix training method based on the training corpus and the feedback information to obtain the superimposed model. The superimposed model is then fused into the preset inference model to obtain the target inference model. The target inference model is then used to adjust the financial problem processing workflow package to obtain a new financial problem processing workflow package.

[0087] In some specific embodiments, the work package generation module 11 may specifically include:

[0088] The analysis result generation unit is used to receive user financial questions issued by users through the input interface in the first loop architecture, and call the problem analysis unit in the preset reasoning model to analyze the user financial questions and obtain analysis results including the essence of the problem and the problem requirements.

[0089] An execution structure generation unit is used to call a structure generation unit connected to the problem analysis unit, so as to use the structure generation unit to generate an execution structure corresponding to the user's financial problem based on the analysis results;

[0090] An execution step generation unit is used to call a step filling unit connected to the structure generation unit and generate specific operation instructions and execution results corresponding to each step in sequence based on the execution order corresponding to the execution structure, thereby obtaining execution steps and generating a financial problem processing workflow package based on the execution structure and the execution steps.

[0091] The work package generation subunit is used to determine the defect information corresponding to the financial problem processing workflow work package, and to call the iterative improvement unit and the feedback loop to adjust, correct and reconstruct the execution structure and execution steps in the financial problem processing workflow work package generated in the previous round based on the defect information, so as to generate a new financial problem processing workflow work package.

[0092] In some specific embodiments, the feedback information generation module 12 may specifically include:

[0093] The first evaluation result generation unit is used to call the structural evaluation model and use preset logical consistency, preset step completeness and preset operability rules to judge the feasibility, coherence and correctness of the execution structure, and obtain a first evaluation result including a first score and structural defect information; the structural defect information is used to describe the unreasonable information of the execution structure.

[0094] The second evaluation result generation unit is used to call the comparative evaluation model to receive each of the financial problem handling process work packages in the same iteration round, and compare the final completion status, result accuracy, process complexity and expression clarity of each of the financial problem handling process work packages to obtain a second evaluation result including a second score and overall defect information; the second score is a relative merit score corresponding to each of the financial problem handling process work packages.

[0095] The third evaluation result generation unit is used to call the execution evaluation model to determine the execution results of the execution steps in each of the financial problem handling process work packages and the connection relationship between the steps, so as to generate a third evaluation result including a third score and alternative solution information based on the execution results and the connection relationship. Then, the target evaluation result is integrated using a preset feedback loop to obtain feedback information. The preset evaluation equipment includes a structural evaluation model, a comparative evaluation model and an execution evaluation model.

[0096] In some specific embodiments, the feedback information generation module 12 may specifically include:

[0097] The execution step determination unit is used to determine whether the execution steps corresponding to each of the financial problem processing workflow packages need to be distinguished by calling external tools, and to obtain the first financial problem processing workflow package that needs to be distinguished by calling external tools and the second financial problem processing workflow package that does not need to be distinguished by calling external tools.

[0098] The correctness assessment unit is used to determine the execution steps and execution records corresponding to each of the first financial problem processing workflow packages, and then use preset evaluation criteria to perform a correctness assessment based on the execution steps and execution records to obtain the corresponding first pending assessment result.

[0099] The result generation unit is used to determine the calling instructions and the return results of the corresponding external tools for each of the second financial problem processing workflow packages, and to evaluate whether the calling instructions and the return results are error-free, whether they meet expectations, and whether the execution time is reasonable, so as to obtain a second pending evaluation result, and to generate a third score based on the first pending evaluation result and the second pending evaluation result.

[0100] An alternative solution information generation unit is used to determine whether the third score is greater than a preset score threshold. If the third score is greater than the preset score threshold, then corresponding alternative solution information is generated based on the third score.

[0101] In some specific embodiments, the training corpus generation module 13 may specifically include:

[0102] The training corpus generation unit is used to call the second loop architecture to format the target evaluation result, obtain the formatted evaluation result, and then use the preset corpus generation model and the training specifications corresponding to the training task to extract, reorganize and standardize the information in the formatted evaluation result, thereby generating a structured training corpus.

[0103] The training corpus storage unit is used to determine the data size corresponding to each training corpus in the preset corpus repository, and to determine whether the data size is greater than a preset size threshold. If the data size is not greater than the preset size threshold, the training corpus is stored in the preset corpus repository using a preset text file system.

[0104] The repository monitoring unit is used to monitor the accumulation status and time status of each training corpus in the preset corpus repository using a preset database system if the data size is not greater than the preset size threshold, and to trigger the model training process when the accumulation status meets a first preset condition or the time status meets a second preset condition.

[0105] In some specific embodiments, the model training module 14 may specifically include:

[0106] The judgment result generation unit is used to call the second loop architecture to judge whether each training corpus in the preset corpus repository is greater than the preset corpus quantity threshold, and obtain the judgment result;

[0107] The superposition model generation unit is used to extract several training corpora from the preset corpus repository if the judgment result indicates that each of the training corpora is greater than the preset corpus quantity threshold, to obtain training samples, and to train and fine-tune the initial large language model using low-rank matrix factorization training technology and based on the training samples to obtain a superposition model with incremental knowledge.

[0108] The inference model determination unit is used to load and fuse the superimposed model into the parameter space of the preset inference model used for analyzing and generating the financial problem processing workflow package to obtain a target inference model, so as to adjust the financial problem processing workflow package using the target inference model to obtain a new financial problem processing workflow package.

[0109] Furthermore, embodiments of this application also disclose an electronic device, Figure 5 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the financial problem processing method based on a dual-loop architecture disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0110] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0111] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0112] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the financial problem processing method based on a dual-loop architecture executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0113] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned disclosed financial problem processing method based on a dual-loop architecture. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0114] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0115] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0116] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0117] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0118] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for processing financial problems based on a double loop architecture, characterized in that, The application is applied to a double-cycle architecture including a first cycle architecture and a second cycle architecture, and includes: calling an input interface in the first cycle architecture to receive a user financial problem, analyzing the user financial problem by using a preset reasoning model, obtaining an analysis result, and generating a plurality of financial problem processing flow work packages including an execution structure and an execution step based on the analysis result; the execution structure is used to describe the sequential flow of solving the user financial problem; the execution step is used to describe the specific operation content and operation result of each flow node; calling the first cycle architecture and a preset evaluation device to respectively perform structural evaluation, execution evaluation and comparative evaluation on each of the financial problem processing flow work packages, obtaining a target evaluation result including a first evaluation result, a second evaluation result and a third evaluation result, and integrating the target evaluation result by using a preset feedback loop to obtain feedback information; calling the second cycle architecture and a preset corpus generation model and processing the target evaluation result based on a preset format to obtain structured training corpus, and storing the training corpus to a preset corpus storage; calling the second cycle architecture to determine whether each of the training corpus in the preset corpus storage meets a preset condition, if so, using a low-rank matrix training method and based on the training corpus and the feedback information to train an initial large language model to obtain a superposition model, and fusing the superposition model into the preset reasoning model to obtain a target reasoning model, so as to adjust the financial problem processing flow work package by using the target reasoning model to obtain a new financial problem processing flow work package.

2. The dual-cycle framework based financial problem processing method of claim 1, wherein, The calling of the input interface in the first cycle architecture to receive a user financial problem, and the analysis of the user financial problem by using a preset reasoning model to obtain an analysis result, and the generation of a plurality of financial problem processing flow work packages including an execution structure and an execution step based on the analysis result, includes: receiving a user financial problem issued by a user by using the input interface in the first cycle architecture, and calling a problem analysis unit in a preset reasoning model to analyze the user financial problem to obtain an analysis result including a problem essence and a problem demand; calling a structure generation unit connected with the problem analysis unit to generate an execution structure corresponding to the user financial problem based on the analysis result by using the structure generation unit; calling a step filling unit connected with the structure generation unit and generating specific operation instructions and execution results corresponding to each step based on the execution order of the execution structure to obtain an execution step, so as to generate a financial problem processing flow work package based on the execution structure and the execution step; determining defect information corresponding to the financial problem processing flow work package, and calling an iterative improvement unit and a feedback loop to adjust, correct and restructure the execution structure and the execution step in the financial problem processing flow work package generated in the last round based on the defect information to generate a new financial problem processing flow work package.

3. The dual-cycle framework based financial problem processing method of claim 1, wherein, The calling the first loop architecture and the preset evaluation device respectively perform structural evaluation, execution evaluation and comparison evaluation on each of the financial problem processing flow work package, obtain a target evaluation result including a first evaluation result, a second evaluation result and a third evaluation result, and integrate the target evaluation result by using a preset feedback loop to obtain feedback information, including: Call the structural evaluation model and use the preset logical self-consistency, preset step completeness and preset operability rules to judge the feasibility, continuity and correctness of the execution structure, and obtain a first evaluation result including a first score and structural defect information; the structural defect information is used to describe the unreasonable information of the execution structure; Call the comparison evaluation model to receive each of the financial problem processing flow work package in the same iteration round, and compare the final completion state, result accuracy, process complexity and expression clarity of each of the financial problem processing flow work package, to obtain a second evaluation result including a second score and overall defect information; the second score is a relative advantage and disadvantage score corresponding to each of the financial problem processing flow work package; Call the execution evaluation model to determine the execution result of the execution step in each of the financial problem processing flow work package and the connection relationship between steps, to generate a third evaluation result including a third score and alternative solution information based on the execution result and the connection relationship, and then integrate the target evaluation result by using a preset feedback loop to obtain feedback information; the preset evaluation device includes a structural evaluation model, a comparison evaluation model and an execution evaluation model.

4. The dual-cycle framework based financial problem processing method of claim 3, wherein, The third evaluation result including the third score and the alternative solution information based on the execution result and the connection relationship, includes: Determine whether the execution step corresponding to each of the financial problem processing flow work package needs to call external tools for distinction, to obtain first financial problem processing flow work packages that need to call external tools for distinction and second financial problem processing flow work packages that do not need to call external tools for distinction; Determine the execution step and the execution record corresponding to each of the first financial problem processing flow work package, and then perform correctness evaluation based on the execution step and the execution record by using a preset evaluation standard, to obtain a corresponding first to-be-processed evaluation result; Determine the call instruction corresponding to each of the second financial problem processing flow work package and the return result of the corresponding external tool, and respectively evaluate whether the call instruction and the return result are error, whether they meet the expectation, and whether the execution time is reasonable, to obtain a second to-be-processed evaluation result, to generate a third score based on the first to-be-processed evaluation result and the second to-be-processed evaluation result; Determine whether the third score is greater than a preset score threshold, if the third score is greater than the preset score threshold, generate corresponding alternative solution information based on the third score.

5. The dual-cycle framework based financial problem processing method of claim 1, wherein, The calling the second loop architecture and the preset corpus generation model and processing the target evaluation result based on a preset format to obtain structured training corpus, and storing the training corpus to a preset corpus storage, includes: The second loop architecture is called to format the target evaluation result to obtain a formatted evaluation result, and then information in the formatted evaluation result is extracted, reorganized and standardized based on a training specification corresponding to a training task by using a preset corpus generation model, so as to generate a structured training corpus; The data size corresponding to each training corpus in the preset corpus storage library is determined, and it is judged whether the data size is greater than a preset size threshold. If the data size is not greater than the preset size threshold, the training corpus is stored in the preset corpus storage library by using a preset text file system; If the data size is not greater than the preset size threshold, the accumulation state and the time state of each training corpus in the preset corpus storage library are monitored by using a preset database system, and when the accumulation state meets a first set condition or the time state meets a second set condition, a model training process is triggered.

6. The dual-cycle framework based financial problem processing method of claim 1, wherein, The second loop architecture is called to determine whether each training corpus in the preset corpus storage library meets a preset condition. If yes, an initial large language model is trained by using a low-rank matrix training method based on the training corpus and the feedback information to obtain a superimposed model, and the superimposed model is fused into the preset reasoning model to obtain a target reasoning model. The target reasoning model is used to adjust the financial problem processing flow work package to obtain a new financial problem processing flow work package, including: The second loop architecture is called to determine whether each training corpus in the preset corpus storage library is greater than a preset corpus quantity threshold to obtain a judgment result. If the judgment result indicates that each training corpus is greater than the preset corpus quantity threshold, a training sample is obtained by extracting a plurality of training corpora from the preset corpus storage library, and an initial large language model is trained and parameter fine-tuned based on the training sample by using a low-rank matrix decomposition training technology to obtain a superimposed model with incremental knowledge. The superimposed model is loaded and fused into the parameter space of the preset reasoning model used for analyzing and generating a financial problem processing flow work package to obtain a target reasoning model, which is used to adjust the financial problem processing flow work package to obtain a new financial problem processing flow work package.

7. A financial problem processing apparatus based on a double loop architecture, characterized by comprising: The double loop architecture includes a first loop architecture and a second loop architecture, including: A work package generation module is configured to call an input interface in the first loop architecture to receive a user financial problem, analyze the user financial problem by using a preset reasoning model to obtain an analysis result, and generate a plurality of financial problem processing flow work packages including execution structures and execution steps based on the analysis result. The execution structure is used to describe the sequential flow of solving the user financial problem, and the execution step is used to describe the specific operation content and operation result of each flow node. The feedback information generation module is configured to call the first loop architecture and a preset evaluation device to perform structural evaluation, execution evaluation, and comparative evaluation on each of the financial problem processing flow work packages, to obtain target evaluation results including a first evaluation result, a second evaluation result, and a third evaluation result, and to integrate the target evaluation results by using a preset feedback loop to obtain feedback information. The training corpus generation module is configured to call the second loop architecture and a preset corpus generation model, and process the target evaluation results based on a preset format to obtain structured training corpora, and store the training corpora in a preset corpus storage. The model training module is configured to call the second loop architecture to determine whether each of the training corpora in the preset corpus storage meets a preset condition, and if so, to perform model training on an initial large language model based on the training corpora and the feedback information by using a low-rank matrix training method to obtain a superimposed model, and to fuse the superimposed model into the preset reasoning model to obtain a target reasoning model, so as to adjust the financial problem processing flow work packages by using the target reasoning model to obtain new financial problem processing flow work packages.

8. The dual-cycle architecture based financial problem processing apparatus according to claim 7, wherein, The feedback information generation module includes: The first evaluation result generation unit is configured to call a structural evaluation model and perform feasibility, coherence, and correctness judgments on the execution structure by using preset logic self-consistency, preset step completeness, and preset operability rules to obtain a first evaluation result including a first score and structural defect information; the structural defect information is used to describe unreasonable information of the execution structure. The second evaluation result generation unit is configured to call a comparative evaluation model to receive each of the financial problem processing flow work packages in the same iteration round, and compare the final completion states, result accuracies, process complexities, and expression clarities of the financial problem processing flow work packages to obtain a second evaluation result including a second score and overall defect information; the second score is a relative advantage and disadvantage score corresponding to each of the financial problem processing flow work packages. The third evaluation result generation unit is configured to call an execution evaluation model to determine execution results of the execution steps in each of the financial problem processing flow work packages and the connection relationships between the steps, to generate a third evaluation result including a third score and alternative solution information based on the execution results and the connection relationships, and to integrate the target evaluation results by using a preset feedback loop to obtain feedback information; the preset evaluation device includes the structural evaluation model, the comparative evaluation model, and the execution evaluation model.

9. An electronic device, comprising: The memory is configured to save a computer program; The processor is configured to execute the computer program to implement the financial problem processing method based on the double-loop architecture as claimed in any one of claims 1 to 6. The memory is configured to save a computer program; and the processor is configured to execute the computer program to implement the financial problem processing method based on the double-loop architecture as claimed in any one of claims 1 to 6.

10. A computer readable storage medium characterized by ​

Images