Method and device for constructing compliance index system based on large language model

Abstract

The invention discloses a compliance index system construction method and device based on a large language model, and relates to the technical field of data processing. The problems that an existing compliance index system is lack of comprehensiveness, insufficient in regulation basis, difficult to update, lack of scene analysis, low in efficiency and the like are solved. Comprising the following steps: when a matching instruction sent by a client is received, obtaining an initial index system plan through a large language model according to an index system conception intention, index system limiting conditions included in an index system instruction, multi-source data and an initial scene classification framework, and sending the initial index system plan to the client, the client confirms whether the indexes included in the initial index system plan are correct or not; and when the audit index system sent by the client is received, determining the final indexes included in the audit index system, the definition of each final index, the calculation method of each final index and the regulation name of each final index, and sending the audit index system to the scene database.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and more specifically to a method and device for constructing a compliance indicator system based on a large language model. Background Art

[0002] The compliance indicator system is a set of systematic and quantitative indicators used to measure an organization's performance in complying with laws and regulations, regulatory requirements, industry standards, internal policies and ethical standards. It sets a series of specific and measurable indicators to assess whether the organization's operations meet relevant compliance requirements. With the continuous improvement of laws and regulations and the continuous improvement of corporate compliance requirements, it is becoming increasingly important to build a scientific and reasonable compliance indicator system.

[0003] Among the existing methods for building compliance indicator systems, some methods mainly rely on expert experience and limited data sources to build indicator systems, which makes it difficult to fully cover all relevant regulatory requirements; although some methods can automatically generate SQL statements to build indicators, they cannot guarantee that each indicator has clear regulatory support, and there are compliance risks; since laws and regulations are often updated and changed, and the update process in the existing indicator system construction methods is relatively complex, there is a problem of difficulty in quickly adapting to regulatory changes; at the same time, most of the existing methods fail to fully consider the particularities of different compliance scenarios, and it is difficult to provide accurate compliance indicators for specific situations. Traditional methods usually require a lot of manpower and time to study regulations and design indicators, and are less efficient; accordingly, many automated methods have difficulty explaining the correspondence between each indicator and specific regulatory provisions, affecting the credibility and practicality of the indicator system.

[0004] To sum up, the existing compliance indicator system lacks comprehensiveness, has insufficient legal basis, is difficult to update, lacks scenario-based analysis, and is inefficient. Summary of the invention

[0005] The embodiments of the present invention provide a method and device for constructing a compliance indicator system based on a large language model, which are used to solve the problems of the existing compliance indicator system, such as lack of comprehensiveness, insufficient legal basis, difficulty in updating, lack of scenario-based analysis, and low efficiency.

[0006] The embodiment of the present invention provides a method for constructing a compliance indicator system based on a large language model, including:

[0007] The processor receives an instruction for constructing an indicator system sent by a client, analyzes the multi-source data included in the instruction for constructing an indicator system based on a large language model, and obtains the inspection scenarios included in the multi-source data, the regulatory name, scenario type, and scenario name of each inspection scenario, wherein the multi-source data is a regulatory document based on which the indicator system is constructed;

[0008] The processor receives the scenario classification concept instruction sent by the client, performs cluster analysis on the scenario classification concept instruction and the inspection scenario based on the large language model, obtains an initial scenario classification framework, and sends the initial scenario classification framework to the client, so that the client confirms whether the initial scenario classification framework matches the indicator system concept intention and indicator system restriction conditions included in the indicator system construction instruction;

[0009] When the processor receives the matching instruction sent by the client, the processor obtains the initial indicator system plan through the large language model according to the indicator system conception intention, the indicator system restriction conditions included in the indicator system construction instruction, the multi-source data and the initial scene classification framework, and sends the initial indicator system plan to the client, so that the client confirms whether the indicators included in the initial indicator system plan are correct;

[0010] When the processor receives the audit indicator system sent by the client, it determines the final indicators included in the audit indicator system, the definition of each final indicator, the calculation method of each final indicator, and the regulatory name of each final indicator, and sends the audit indicator system to the scenario database.

[0011] Preferably, the large language model analyzes the multi-source data included in the instruction for building the indicator system, specifically including:

[0012] The processor identifies the regulatory documents included in the multi-source data, and confirms the compliance check scenario involved in each regulation included in the regulatory documents;

[0013] Confirm the scenario type, scenario name, and regulatory name and specific clauses corresponding to each inspection scenario.

[0014] Preferably, the cluster analysis of the scene classification conception instructions and the inspection scenes based on the large language model to obtain an initial scene classification framework specifically includes:

[0015] The processor parses the scene classification conception instruction to obtain at least a desired scene classification method, converts the text of the inspection scene into a vector representation in a high-dimensional vector space, and in the vector space, sequentially determines the similarity of scene types included in the inspection scene according to the similarity between the vectors;

[0016] Classifying the multiple scene types whose scene type similarity is greater than a first threshold into the same category, and performing hierarchical division according to the scene type similarities of the multiple scene types included in the same category to obtain initial hierarchical clustering;

[0017] According to the scene classification method, the initial hierarchical clustering is adjusted, and semantic labels are determined for the category assignments included in the updated hierarchical clustering to obtain the initial scene classification framework; the semantic labels are used to summarize the core features of the inspection scenes included in the same category.

[0018] Preferably, the initial indicator system planning is obtained through a large language model according to the indicator system conception intention, the indicator system restriction conditions included in the indicator system construction instruction, the multi-source data and the initial scene classification framework, specifically including:

[0019] Determine the characteristics of each type of scene included in the initial scene classification framework, extract the corresponding key evaluation factors of each type of scene, and design indicators based on the key evaluation factors;

[0020] Determine the calculation method of the indicator according to the conception intention of the indicator system and the restriction conditions of the indicator system;

[0021] Determine the data source of the indicator according to the multi-source data, and establish the hierarchical relationship and association relationship between the indicators;

[0022] The initial indicator system planning including the definition of indicators, the calculation method of indicators and the data source of indicators is formed according to the hierarchical relationship and the association relationship.

[0023] Preferably, the inspection scenarios included in the obtained multi-source data, after the regulation name, scenario type and scenario name of each inspection scenario, further include:

[0024] The inspection scenarios included in the multi-source data, the regulatory name of each inspection scenario, the scene type of each inspection scenario and the scene name of each inspection scenario are sent to a scenario database, so that the scenario database stores the inspection scenarios, the regulatory name of each inspection scenario, the scene type of each inspection scenario and the scene name of each inspection scenario.

[0025] The embodiment of the present invention provides a method for constructing a compliance indicator system based on a large language model, including:

[0026] Sending an indicator system construction instruction to the processor, so that the processor analyzes the multi-source data included in the indicator system construction instruction to obtain the inspection scenarios included in the multi-source data, and the regulatory name, scenario type and scenario name of each inspection scenario;

[0027] Sending a scene classification conception instruction to a processor, so that the processor performs cluster analysis on the scene classification conception instruction and the inspection scene and obtains an initial scene classification framework;

[0028] The initial scene classification framework sent by the processor is received, and when it is confirmed that the initial scene classification framework matches the indicator system conception intention and the indicator system restriction condition included in the indicator system construction instruction, a matching instruction is sent to the processor, so that the processor obtains an initial indicator system plan through a large language model according to the indicator system conception intention, the indicator system restriction condition included in the indicator system construction instruction, the multi-source data and the initial scene classification framework;

[0029] The initial indicator system plan sent by the processor is received, and when it is confirmed that the indicators included in the initial indicator system plan are correct, the audit indicator system is sent to the processor, so that the processor can determine the final indicators included in the audit indicator system, the definition of each of the final indicators, the calculation method of each of the final indicators, and the regulatory name of each of the final indicators.

[0030] The embodiment of the present invention provides a device for constructing a compliance indicator system based on a large language model, including:

[0031] An analysis unit, configured to receive an instruction for constructing an indicator system sent by a client, analyze the multi-source data included in the instruction for constructing an indicator system based on a large language model, and obtain the inspection scenarios included in the multi-source data, the regulatory name, scenario type, and scenario name of each inspection scenario, wherein the multi-source data is a regulatory document based on which the indicator system is constructed;

[0032] A first obtaining unit is configured to receive a scenario classification concept instruction sent by a client, perform cluster analysis on the scenario classification concept instruction and the inspection scenario based on a large language model, obtain an initial scenario classification framework, and send the initial scenario classification framework to the client, so that the client can confirm whether the initial scenario classification framework matches the indicator system concept intention and indicator system restriction conditions included in the indicator system construction instruction;

[0033] A second obtaining unit is configured to obtain an initial indicator system plan through a large language model according to the indicator system conception intention, the indicator system restriction conditions included in the indicator system construction instruction, the multi-source data and the initial scene classification framework when the processor receives the matching instruction sent by the client, and send the initial indicator system plan to the client to enable the client to confirm whether the indicators included in the initial indicator system plan are correct;

[0034] A determination unit is used to determine the final indicators included in the audit indicator system, the definition of each final indicator, the calculation method of each final indicator, and the regulatory name of each final indicator when the processor receives the audit indicator system sent by the client, and send the audit indicator system to the scenario database.

[0035] The embodiment of the present invention provides a device for constructing a compliance indicator system based on a large language model, including:

[0036] A first sending unit is used to send an indicator system construction instruction to the processor, so that the processor analyzes the multi-source data included in the indicator system construction instruction to obtain the inspection scenarios included in the multi-source data, and the regulatory name, scenario type and scenario name of each inspection scenario;

[0037] A second sending unit is used to send a scene classification conception instruction to the processor, so that the processor performs cluster analysis on the scene classification conception instruction and the inspection scene and obtains an initial scene classification framework;

[0038] A judgment unit, configured to receive the initial scene classification framework sent by the processor, and when confirming that the initial scene classification framework matches the indicator system conception intention and indicator system restriction conditions included in the indicator system construction instruction, send a matching instruction to the processor, so that the processor obtains an initial indicator system plan through a large language model according to the indicator system conception intention, the indicator system restriction conditions included in the indicator system construction instruction, the multi-source data and the initial scene classification framework;

[0039] The audit unit is used to receive the initial indicator system plan sent by the processor, and when it is confirmed that the indicators included in the initial indicator system plan are correct, send the audit indicator system to the processor, so that the processor can determine the final indicators included in the audit indicator system, the definition of each final indicator, the calculation method of each final indicator, and the regulatory name of each final indicator.

[0040] An embodiment of the present invention provides a computer device, comprising a scenario database and a processor, wherein the scenario database stores a computer program, and when the computer program is executed by the processor, the processor executes any one of the above-mentioned methods for constructing a compliance indicator system based on a large language model.

[0041] An embodiment of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes any one of the above-mentioned methods for constructing a compliance indicator system based on a large language model.

[0042] The method for constructing a compliance indicator system based on a large language model provided in an embodiment of the present invention can comprehensively analyze all relevant laws and regulations through the powerful text understanding ability of the large language model to avoid missing important compliance requirements; the method has a high degree of automation, which can greatly reduce the time of manual analysis and design, and significantly improve the efficiency of indicator system construction; the indicator system generated by it through scene recognition and clustering is more in line with actual application needs; each indicator can be traced back to specific regulatory provisions, which improves the credibility and practicality of the indicator system; it can generate a customized indicator system according to different construction intentions and indicator system constraints; further, when the regulations change, the indicator system can be quickly updated to maintain consistency with the latest regulations; the method combines the efficiency of the large language model and the judgment of the client to ensure the quality and applicability of the indicator system. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 A schematic diagram of a processor-side flow chart in a method for constructing a compliance indicator system based on a large language model provided in an embodiment of the present invention;

[0045] Figure 2 A schematic diagram of the client process in the method for constructing a compliance indicator system based on a large language model provided in an embodiment of the present invention;

[0046] Figure 3 A schematic diagram of a flow chart of a method for constructing a compliance indicator system based on a large language model provided in an embodiment of the present invention;

[0047] Figure 4 A schematic diagram of the structure of a processor end in a device for constructing a compliance indicator system based on a large language model provided in an embodiment of the present invention;

[0048] Figure 5 A schematic diagram of the client structure in the device for building a compliance indicator system based on a large language model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] The method for constructing a compliance indicator system based on a large language model provided in the embodiment of the present invention aims to solve the problems of the indicator system in the prior art, such as lack of comprehensiveness, insufficient legal basis, difficulty in updating, lack of scenario analysis, low efficiency, and lack of interpretability. The method can quickly and comprehensively analyze relevant laws and regulations, automatically generate a highly targeted and well-interpretable compliance indicator system, and significantly improve the construction efficiency and quality.

[0051] Terms used in this document:

[0052] 1. Large Language Model (LLM) is an artificial intelligence model based on deep learning technology, which usually has a large number of parameters and powerful language understanding and generation capabilities. It can learn language grammar, semantics, pragmatics and other knowledge through unsupervised or self-supervised learning on large-scale text data, so as to complete a variety of natural language processing tasks, such as text generation, knowledge question answering, reasoning calculation, reading comprehension, etc.

[0053] In order to provide a method for constructing a compliance indicator system, the embodiment of the present invention provides a method and device for constructing a compliance indicator system based on a large language model. The preferred embodiments of the present invention are described below in conjunction with the specification and the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention. In addition, the embodiments in this application and the features in the embodiments can be combined with each other if there is no conflict.

[0054] The embodiment of the present invention provides a flowchart of a method for constructing a compliance indicator system based on a large language model on the processor side, see Figure 1 As shown, it mainly includes the following steps:

[0055] Step 101, the processor receives an instruction for constructing an indicator system sent by the client, and analyzes the multi-source data included in the instruction for constructing the indicator system based on a large language model to obtain the inspection scenarios included in the multi-source data, the regulatory name, scenario type and scenario name of each inspection scenario, wherein the multi-source data is the regulatory document on which the indicator system is constructed.

[0056] In step 102, the processor receives a scenario classification concept instruction sent by the client, performs cluster analysis on the scenario classification concept instruction and the inspection scenario based on a large language model, obtains an initial scenario classification framework, and sends the initial scenario classification framework to the client so that the client can confirm whether the initial scenario classification framework matches the indicator system concept intention and indicator system restriction conditions included in the indicator system construction instruction.

[0057] Step 103, when the processor receives the matching instruction sent by the client, it obtains the initial indicator system plan through the large language model according to the indicator system conception intention, the indicator system restriction conditions included in the indicator system construction instruction, the multi-source data and the initial scene classification framework, and sends the initial indicator system plan to the client to enable the client to confirm whether the indicators included in the initial indicator system plan are correct.

[0058] Step 104, when the processor receives the audit indicator system sent by the client, it determines the final indicators included in the audit indicator system, the definition of each final indicator, the calculation method of each final indicator, and the regulatory name of each final indicator, and sends the audit indicator system to the scenario database.

[0059] In step 101, the processor receives an instruction to construct an indicator system sent by the client. Generally, before sending the instruction to construct the indicator system, the client will first obtain the relevant laws and regulations required to build the indicator system, as well as the intention of constructing the indicator system and the restrictions on the indicator system. For example, an e-commerce platform wants to build a commodity sales compliance indicator system. The relevant legal texts may include the "Consumer Protection Law of the People's Republic of China" and the "Product Quality Law of the People's Republic of China". The intention of construction is to ensure that the commodity sales activities on the platform are legal and compliant and to protect the rights and interests of consumers; the restriction of the indicator system is that the indicator system must be able to collect and calculate data based on the existing platform data system.

[0060] Furthermore, the processor parses the received instruction for constructing the indicator system to obtain multi-source data, the indicator system conception and intention, and the indicator system restriction conditions.

[0061] In an embodiment of the present invention, after determining the multi-source data included in the instruction for building an indicator system, the multi-source data can be input into a large language model; the large language model identifies the regulatory documents included in the multi-source data based on the received multi-source data and the recognition task requirements, and confirms the compliance inspection scenarios involved in each regulation included in the regulatory documents; confirms the scenario type and scenario name of each inspection scenario, as well as the regulatory name and specific clauses corresponding to each inspection scenario.

[0062] For example, when the specific provisions of the "Product Quality Law of the People's Republic of China" are input into the GPT-4 (Generative Pretrained Transformer4) large language model, in order for the large language model to understand and analyze the content of the regulations, the recognition task "Please identify the compliance inspection scenarios involved in the "Product Quality Law of the People's Republic of China"" will be issued to GPT-4 to guide the large language model to conduct targeted analysis. The large language model identifies a scenario that "the products sold must have a quality inspection certificate", the scenario type is "must be implemented", the scenario name is "product quality inspection certificate provision", the corresponding regulation name is "Product Quality Law of the People's Republic of China", and the specific clause is "Article 27 The label on the product or its packaging must be authentic and meet the following requirements: (a) There is a product quality inspection certificate."

[0063] In the embodiment of the present invention, after the large language model analyzes the input multi-source data, the analysis results are sent to the scene database for storage.

[0064] In step 102, when the processor receives the scene classification concept instruction sent by the client again, all the inspection scenes and scene classification concept instructions obtained in step 101 are input into the large language model, and the large language model is instructed to perform cluster analysis on all the inspection scenes, and the initial scene classification framework is obtained according to the cluster analysis results.

[0065] Specifically, the scene classification concept instruction is parsed to obtain at least the desired scene classification method, the text of the inspection scene is converted into a vector representation in a high-dimensional vector space, and in the vector space, the scene type similarities included in the inspection scene are determined in turn according to the similarities between the vectors; multiple scene types whose scene type similarities are greater than a first threshold are classified into the same category, and hierarchical division is performed according to the scene type similarities of the multiple scene types included in the same category to obtain initial hierarchical clustering; according to the scene classification method, the initial hierarchical clustering is adjusted to obtain updated hierarchical clustering, and then semantic labels are assigned to the update categories included in the updated hierarchical clustering to obtain an initial scene classification framework; the semantic labels are used to summarize the core features of the inspection scenes included in the update categories.

[0066] For example, suppose a medical insurance company wants to build a compliance indicator system. In the above steps, a series of compliance inspection scenarios are identified, such as "verification of customer information authenticity", "complaint process compliance review", "insurance rate calculation compliance check", etc. In this step, the scenario classification concept instruction "classify according to business process stage" sent by the client is received.

[0067] The large language model calculates the semantic similarity between the instruction "classify by business process stage" and each inspection scenario. For example, the scenario of "verifying the authenticity of customer information" may have a high semantic similarity with the "information collection stage at the beginning of the business", and the scenario of "compliance review of the claims process" may have a high semantic similarity with the "claims processing stage at the end of the business".

[0068] Assuming the first threshold is 0.6, the large language model selects inspection scenarios with semantic similarity higher than 0.6 for preliminary clustering. Scenarios such as "verification of customer information authenticity" and "compliance inspection of insurance contract signing" are classified as "early stage of business"; scenarios such as "compliance inspection of insurance service provision" and "compliance review of customer complaint handling" are classified as "mid-stage of business"; scenarios such as "compliance review of claims process" and "compliance inspection of insurance premium settlement" are classified as "late stage of business".

[0069] Furthermore, the hierarchical clustering algorithm is used to subdivide the preliminary clustering results. For example, in the "early stage of business" category, the scenarios are further divided into "customer information collection sub-stage" and "contract signing sub-stage"; in the "mid-stage of business" category, the scenarios are divided into "service provision sub-stage" and "complaint handling sub-stage".

[0070] According to the instruction of "classify by business process stage", check and adjust the hierarchical clustering results. If it is found that "insurance premium calculation compliance check" was originally mistakenly classified as "late business stage", but it is actually more in line with "early business rate setting stage", then adjust it to the corresponding category to obtain updated hierarchical clustering. Assign semantic labels to the updated categories included in the updated hierarchical clustering. For example, "early business-customer information collection sub-stage", "mid-business-service provision sub-stage", "late business-claims processing sub-stage", etc.

[0071] Finally, an initial scenario classification framework based on business process stages is formed, including the early business stage, mid-business stage, and late business stage.

[0072] 1) In the early stage of business, the customer information collection sub-stage: verification of customer information authenticity, compliance check of customer health information collection, etc.; the contract signing sub-stage: compliance check of insurance contract signing, compliance review of terms and conditions notification, etc.

[0073] 2) Mid-term business stage: service provision sub-stage: compliance inspection of insurance service provision, compliance review of service quality supervision, etc.; complaint handling sub-stage: compliance review of customer complaint handling, compliance inspection of complaint response time, etc.

[0074] 3) In the later stage of business, the claims processing sub-stage: compliance review of claims process, compliance review of claims amount calculation, etc.; the rate settlement sub-stage: compliance review of insurance premium settlement, compliance review of rate adjustment, etc.

[0075] In this step, the obtained initial scene classification framework is sent to the client, so that the client can confirm whether the initial scene classification framework matches the indicator system conception intention and indicator system restriction conditions included in the instruction to build the indicator system.

[0076] It should be noted that, in the embodiment of the present invention, the specific range of the first threshold is not limited.

[0077] In step 103, when the processor receives the matching instruction sent by the client, it can be considered that the client has determined that the initial scene classification framework matches the indicator system conception intention and indicator system restriction conditions included in the indicator system construction instruction. Further, it is necessary to obtain the initial indicator system plan through a large language model based on the indicator system conception intention, the indicator system restriction conditions included in the indicator system construction instruction, the multi-source data and the initial scene classification framework.

[0078] In an embodiment of the present invention, a large language model is used to generate a preliminary initial indicator system plan. In simple terms, the indicator system concept intention, indicator system constraints included in the indicator system construction instructions, multi-source data and initial scenario classification framework are input into the large language model so that the large language model can design an indicator system that meets the requirements and generate an initial indicator system plan that includes information such as the definition of indicators, the calculation method of indicators and the data source of indicators.

[0079] Specifically, the large language model determines the characteristics of each type of scenario included in the initial scenario classification framework, extracts the corresponding key evaluation factors of each type of scenario, and designs indicators based on the key evaluation factors. Assume that an Internet financial company wants to build a compliance indicator system. After the previous steps, it obtains a scenario classification scheme that has passed the review. If there is a type of scenario in the scenario classification scheme that is "user loan information review compliance", the characteristics of this type of scenario include the completeness of the review process, the consistency of the review standards, the accuracy of the review information, etc. The key evaluation factors extracted are: the completeness of the review process steps, the deviation rate of the implementation of the review standards, and the accuracy rate of the verification of the loan information. Further, the indicator "completion rate of the review process steps" can be designed for the "completeness of the review process steps"; the indicator "deviation rate of the review standards" can be designed for the "deviation rate of the implementation of the review standards"; the indicator "accurate proportion of the verification of the loan information" can be designed for the "accuracy rate of the verification of the loan information".

[0080] Furthermore, the calculation method of the indicator is determined according to the conceptual intent of the indicator system and the constraint conditions of the indicator system. Combining the above example, it can be seen that the conceptual intent of the indicator system is to ensure the compliance of the lending business, and the constraint condition of the indicator system is that the data can be obtained from the company's internal system.

[0081] How to calculate the “Audit process step completion rate”: ”.

[0082] Calculation method of “Audit Standard Deviation Rate”: ”.

[0083] Calculation method of “Accurate ratio of loan information verification”: ”.

[0084] Furthermore, the data sources of the indicators are determined based on multi-source data, and the hierarchical relationship and association relationship between the indicators are established; combined with the above examples, it can be seen that the data source of the "audit process step completion rate" is: the company's internal loan audit system records, which will record the execution of the audit steps of each loan application. The data source of the "audit standard deviation rate" is: the audit result record and the standard audit rule document, and the deviation is determined by comparing the two. The data source of the "loan information verification accuracy ratio" is: the loan information verification report and the original loan information. The verification report is provided by a special verification department.

[0085] Hierarchical relationship: "Audit process step completion rate", "Audit standard deviation rate" and "Accurate proportion of loan information verification" are all subordinate to the superior indicator of "User loan information review compliance".

[0086] Correlation relationship: A low "audit process step completion rate" may lead to an increase in the "audit standard deviation rate", because an incomplete audit process may not be able to accurately implement the audit standards; a low "loan information verification accuracy ratio" may also affect the "audit standard deviation rate", and inaccurate loan information may cause the audit results to deviate from the standard.

[0087] Furthermore, an initial indicator system planning including the definition of indicators, the calculation method of indicators and the data source of indicators is formed according to the hierarchical relationship and the association relationship. Combining the above examples, the initial indicator system planning shown in Table 1 can be obtained:

[0088] Table 1: Initial indicator system planning

[0089]

[0090] The initial indicator system planning provided by the embodiment of the present invention will serve as the basis for subsequent client audit and optimization. Through continuous adjustment and improvement, an audit indicator system that meets the needs will eventually be formed.

[0091] In step 103, after the processor obtains the initial indicator system plan based on the large language model, it needs to send the initial indicator system plan to the client, so that the client can confirm whether the indicators included in the initial indicator system plan are correct.

[0092] In step 104, when the processor receives the audit indicator system sent by the client, it can be considered that the client has determined that the initial indicator system plan or the modified indicator system plan formed in the above steps has been confirmed by the client. At this time, the processor needs to determine the final indicators included in the audit indicator system, the definition of each final indicator, the calculation method of each final indicator, and the regulatory name of each final indicator. Then the audit indicator system is sent to the scenario database.

[0093] The method for constructing a compliance indicator system based on a large language model provided by an embodiment of the present invention can analyze multimodality, that is, in addition to text, it can also analyze multimodal data such as charts and audio to obtain more comprehensive information; it can integrate external knowledge bases, that is, combine professional knowledge bases with large language models to improve professionalism and accuracy; it can dynamically adjust weights, that is, dynamically adjust the weights of different indicators according to actual application effects; it can support cross-language, that is, support multilingual regulatory analysis, and is suitable for the compliance needs of multinational companies; it can analyze indicator dependencies, that is, automatically identify and display the dependencies between indicators to form an indicator network diagram; it can provide compliance risk warning, that is, develop compliance risk warning functions based on the indicator system; it can perform historical version management, that is, record the evolution history of the indicator system, and support version backtracking and comparison. By adopting the method provided by an embodiment of the present invention, the construction efficiency and quality of the compliance indicator system can be significantly improved. This method can not only fully cover the requirements of relevant regulations, but also generate highly targeted indicators according to specific application scenarios. At the same time, since each indicator is directly related to specific regulatory provisions, the interpretability and credibility of the indicator system are greatly improved. In addition, this method has a high degree of automation and can quickly respond to regulatory changes and update the indicator system in a timely manner, effectively reducing compliance risks.

[0094] Compared with the prior art, the method for constructing a compliance indicator system based on a large language model provided in an embodiment of the present invention can comprehensively analyze all relevant laws and regulations through the powerful text understanding ability of the large language model to avoid missing important compliance requirements; the method has a high degree of automation, which can greatly reduce the time for manual analysis and design, and significantly improve the efficiency of indicator system construction; the indicator system generated by it through scene recognition and clustering is more in line with actual application needs; each indicator can be traced back to specific regulatory provisions, which improves the credibility and practicality of the indicator system; it can generate a customized indicator system according to different construction intentions and indicator system constraints; further, when the regulations change, the indicator system can be quickly updated to maintain consistency with the latest regulations; the method combines the efficiency of the large language model and the judgment of the client to ensure the quality and applicability of the indicator system.

[0095] Based on the same inventive concept, the embodiment of the present invention provides a method flow chart of the client in the method for building a compliance indicator system based on a large language model, see Figure 2 As shown, it mainly includes the following steps:

[0096] In step 201, the client sends an instruction to construct an indicator system to the processor, so that the processor analyzes the multi-source data included in the instruction to construct the indicator system and obtains the inspection scenarios included in the multi-source data, the regulatory name, scenario type and scenario name of each inspection scenario.

[0097] Step 202: The client sends a scene classification concept instruction to the processor, so that the processor performs cluster analysis on the scene classification concept instruction and the inspection scene and obtains an initial scene classification framework.

[0098] In step 203, the client receives the initial scene classification framework sent by the processor. When it is confirmed that the initial scene classification framework matches the indicator system concept intention and indicator system constraints included in the indicator system construction instruction, a matching instruction is sent to the processor so that the processor obtains the initial indicator system plan according to the indicator system concept intention, the indicator system constraints included in the indicator system construction instruction, the multi-source data and the initial scene classification framework through a large language model.

[0099] In step 204, the client receives the initial indicator system plan sent by the processor. When confirming that the indicators included in the initial indicator system plan are correct, the audit indicator system is sent to the processor so that the processor can determine the final indicators included in the audit indicator system, the definition of each final indicator, the calculation method of each final indicator, and the regulatory name of each final indicator.

[0100] In step 201, after collecting the multi-source data required for constructing the indicator system, the client generates an indicator system construction instruction according to the indicator system construction intention, indicator system restriction conditions and the multi-source data, and sends the indicator system construction instruction to the processor.

[0101] Furthermore, the client sends an instruction to construct an indicator system to the processor, so that the processor can analyze the multi-source data included in the instruction to construct the indicator system, and obtain the inspection scenarios, the regulatory name, scenario type and scenario name of each inspection scenario based on the multi-source data.

[0102] In step 202, after the client sends the instruction to construct the indicator system to the processor, it sends the scene classification concept instruction to the processor again, so that the processor can perform cluster analysis on the scene classification concept instruction and the inspection scene and obtain an initial scene classification framework.

[0103] In step 203, the client receives the initial scene classification framework sent by the processor and needs to confirm whether the initial scene classification framework matches the indicator system conception intention and indicator system restriction conditions included in the indicator system construction instruction, including:

[0104] The initial scene classification framework generated by the large language model is presented on a display screen or in other ways to facilitate the review of the initial scene classification framework by the reviewer. The reviewer is usually a reviewer who is familiar with business rules, regulatory requirements or has a deep understanding of compliance, such as the company's compliance managers, business experts, etc. The reviewer can evaluate the initial scene classification framework from a practical business perspective. Furthermore, the client can also collect the reviewer's opinions and suggestions on the initial scene classification framework. For example, the reviewer may find unreasonable aspects in the initial scene classification framework, such as inaccurate scene classification, unclear category division, omission or incorrect classification of certain scenes, and make corresponding modification suggestions;

[0105] If the client receives modification suggestions from the reviewer, it can send the modification suggestions to the processor. The purpose is to allow the large language model in the processor to adjust the initial scene classification framework according to the modification suggestions and generate a modified scene classification framework. After the client receives the modified scene classification framework from the processor, it will be presented again on a display screen or in other ways until the reviewer is satisfied with the modified scene classification framework and believes that the solution is accurate, reasonable and in line with business realities.

[0106] Furthermore, the client sends a matching instruction to the processor, so that the processor can obtain the initial indicator system plan through the large language model according to the indicator system conception intention, the indicator system constraints included in the indicator system construction instruction, the multi-source data and the initial scene classification framework (modify the scene classification framework).

[0107] For example, suppose an online education platform wants to build a compliance indicator system. After completing step 202 above, it receives the initial scenario classification framework shown in Table 2:

[0108] Table 2: Initial scene classification framework

[0109]

[0110] The client displays the above initial scene classification framework through a display screen or other means, and the reviewer on the review end reviews the above initial scene classification framework. After review, the review end proposes the following modification suggestions:

[0111] 1) The scenario of “course content complies with the requirements of the education syllabus” should be added to the “course content compliance category” because the courses on online education platforms need to follow the relevant education syllabus.

[0112] 2) "User information management category" can be further subdivided into two subcategories: "User registration information management" and "User privacy information management" to make the classification clearer.

[0113] 3) The “Advertising compliance category” should be supplemented with the “Advertising frequency compliance” scenario to avoid excessive advertising causing trouble to users.

[0114] The client receives the above modification suggestions and sends them to the processor. The large language model re-cluster the initial scene classification framework based on the modification suggestions, that is, feeds these modification suggestions back to the large language model (such as GPT-4), and re-enters all scenes and modification requirements to instruct the large language model to re-cluster the scenes. The large language model generates a modified scene classification framework as shown in Table 3 based on the new requirements:

[0115] Table 3: Modified scene classification framework

[0116]

[0117] The client repeats the above process until it receives feedback from the review end that the review has passed.

[0118] In step 204, after receiving the initial indicator system plan sent by the processor, the client needs to confirm whether the indicators therein are correct, including:

[0119] The initial indicator system plan generated by the large language model, including the definition of indicators, the calculation method of indicators and the data source of indicators, is displayed on a display screen or other means, so that auditors on the audit end can comprehensively review the indicator system design plan from different angles.

[0120] If the auditor finds that there are many problems in the initial indicator system planning, such as some indicators cannot accurately reflect the business compliance, the calculation method is too complex to implement, the data source is unstable or difficult to obtain, and the logical relationship between indicators is unclear, etc., then he can put forward targeted modification suggestions based on his own professional knowledge and actual work experience.

[0121] When the client receives the modification opinions sent by the audit end, it sends the modification opinions to the processor, and the large language model generates a modified indicator system plan based on the modification opinions; that is, these modification opinions are fed back to the large language model, and the "indicator system concept intention", indicator system constraints, scenario classification scheme, and original regulatory text and other related information are input again, and the large language model is instructed to regenerate the modified indicator system plan based on the new requirements. Repeat the above steps until the client receives the approval feedback from the audit end. The client sends the audit indicator system to the processor. On the one hand, the processor confirms the final indicators included in the audit indicator system, the definition of each final indicator, the calculation method of each final indicator, and the regulatory name of each final indicator, and sends the above audit indicator system to the scenario database through the processor for storage.

[0122] For example, suppose a company engaged in cross-border e-commerce business obtains the initial indicator system planning shown in Table 4:

[0123] Table 4: Initial indicator system planning

[0124]

[0125] Display the initial indicator system plan to users (clients): The company will display the above initial indicator system plan to the cross-border business department, legal department and other relevant departments for review.

[0126] Received modification opinions: After careful review, the cross-border business department proposed that the calculation method of the "product compliance and timely listing rate" should take into account the differences in the difficulty of listing different categories of goods, set different weights for different categories, and make the indicator more reasonable; the legal department pointed out that in the "cross-border logistics violation rate", the definition of logistics violations should be clearer, and it needs to be refined to specific types of violations, such as customs clearance violations, transportation overtime violations, etc.; the financial department believes that the data source of the "customer complaint handling compliance rate" only relies on the customer service management system, which may have incomplete data problems. It should be combined with the cost data related to complaint handling in the financial system to comprehensively judge whether the complaint handling is compliant.

[0127] The modification opinions are sent to the processor, and the large language model regenerates the modification index system plan based on the modification opinions to obtain the modification index system plan shown in Table 5:

[0128] Table 5: Modification of indicator system planning

[0129]

[0130] The processor repeats the above process until it passes the audit end review, that is, the above multiple departments believe that the revised indicator system plan is more complete and can accurately reflect the compliance of the company's cross-border e-commerce business. The processor sends the audit indicator system that finally passes the review to the processor, and the audit indicator system will serve as the basis for the final construction of the compliance indicator system. At the same time, the audit indicator system is also stored in the scenario database through the processor.

[0131] Based on the same inventive concept, the embodiment of the present invention also provides a method for constructing a compliance indicator system based on a large language model, such as Figure 3 As shown, the method comprises the following steps:

[0132] Step 301: The client completes the construction of the indicator system and sends an instruction to the processor to construct the indicator system.

[0133] In step 302, the processor analyzes the multi-source data included in the instruction for constructing the indicator system based on the large language model, and obtains the inspection scenarios included in the multi-source data, and the regulatory name, scenario type and scenario name of each inspection scenario.

[0134] Step 303: Send the regulatory name, scenario type and scenario name of the inspection scenario to the scenario database, which stores each received inspection scenario, the regulatory name, scenario type and scenario name of each inspection scenario.

[0135] Step 304: The client sends a scene classification conception instruction to the processor.

[0136] Step 305 , the processor performs cluster analysis on the scene classification conception instructions and the inspection scenes based on the large language model according to the scene classification conception instructions, and obtains an initial scene classification framework.

[0137] Step 306: The processor sends the initial scene classification framework to the client.

[0138] Step 307, the client receives the initial scene classification framework, confirms whether the initial scene classification framework matches the conceptual intent of the indicator system and the constraint conditions of the indicator system, and sends a matching instruction to the processor if it matches; if it does not match, sends a modification instruction to the client.

[0139] Step 308-1: When the processor receives the modification instruction sent by the client, it modifies the initial scene classification framework according to the modification opinion, obtains a modified scene classification framework, and sends the modified scene classification framework to the client.

[0140] Step 308-2, when the processor receives the matching instruction sent by the client, it obtains the initial indicator system plan through the large language model according to the indicator system conception intention, the indicator system constraints included in the indicator system construction instruction, the multi-source data and the initial scene classification framework, and sends the initial indicator system plan to the client.

[0141] Step 309, the client receives the initial indicator system plan, confirms whether the initial scenario classification framework matches the indicator system conception intention and indicator system restriction conditions, and if so, sends the review indicator system to the processor; if not, sends modification suggestions to the processor.

[0142] Step 310 - 1 : When the processor receives the modification suggestion sent by the client, it modifies the initial indicator system plan according to the modification suggestion, obtains the modified indicator system plan, and sends the modified indicator system plan to the client.

[0143] Step 310 - 2 , when the processor receives the audit indicator system sent by the client, it determines the final indicators included in the audit indicator system, the definition of each final indicator, the calculation method of each final indicator and the regulatory name of each final indicator.

[0144] In step 311, the processor sends the final indicators included in the audit indicator system, the definition of each final indicator, the calculation method of each final indicator, and the regulatory name of each final indicator to the scenario database.

[0145] Based on the same inventive concept, an embodiment of the present invention provides a device for constructing a compliance indicator system based on a large language model. Since the principle of the device for solving the technical problem is similar to the method for constructing a compliance indicator system based on a large language model, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0146] like Figure 4 As shown, the device includes an analyzing unit 401 , a first obtaining unit 402 , a second obtaining unit 403 and a determining unit 404 .

[0147] The analysis unit 401 is used for the processor to receive the instruction for building the indicator system sent by the client, analyze the multi-source data included in the instruction for building the indicator system based on the large language model, and obtain the inspection scenarios included in the multi-source data, the regulatory name, scenario type and scenario name of each inspection scenario, wherein the multi-source data is the regulatory document on which the indicator system is built.

[0148] The first obtaining unit 402 is used for the processor to receive the scene classification concept instruction sent by the client, perform cluster analysis on the scene classification concept instruction and the inspection scene based on the large language model, obtain an initial scene classification framework, and send the initial scene classification framework to the client to enable the client to confirm whether the initial scene classification framework matches the indicator system concept intention and indicator system restriction conditions included in the indicator system construction instruction.

[0149] The second obtaining unit 403 is used to obtain an initial indicator system plan through a large language model according to the indicator system conception intention, the indicator system restriction conditions included in the indicator system construction instruction, the multi-source data and the initial scene classification framework when the processor receives a matching instruction sent by the client, and send the initial indicator system plan to the client to enable the client to confirm whether the indicators included in the initial indicator system plan are correct.

[0150] The determination unit 404 is used to determine the final indicators included in the audit indicator system, the definition of each final indicator, the calculation method of each final indicator, and the regulatory name of each final indicator when the processor receives the audit indicator system sent by the client, and send the audit indicator system to the scenario database.

[0151] Based on the same inventive concept, an embodiment of the present invention provides a device for constructing a compliance indicator system based on a large language model. Since the principle of the device for solving the technical problem is similar to the method for constructing a compliance indicator system based on a large language model, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0152] like Figure 5 As shown, the device includes a first sending unit 501, a second sending unit 502, a judging unit 503 and a reviewing unit 504.

[0153] The first sending unit 501 is used to send an instruction to build an indicator system to the processor, so that the processor analyzes the multi-source data included in the instruction to build the indicator system, obtains the inspection scenarios included in the multi-source data, and the regulatory name, scenario type and scenario name of each inspection scenario.

[0154] The second sending unit 502 is used to send a scene classification concept instruction to the processor, so that the processor performs cluster analysis on the scene classification concept instruction and the inspection scene and obtains an initial scene classification framework.

[0155] The judgment unit 503 is used to receive the initial scene classification framework sent by the processor, and when it is confirmed that the initial scene classification framework matches the indicator system concept intention and indicator system restriction conditions included in the indicator system construction instruction, send a matching instruction to the processor, so that the processor obtains the initial indicator system plan through a large language model according to the indicator system concept intention, the indicator system restriction conditions included in the indicator system construction instruction, the multi-source data and the initial scene classification framework.

[0156] The audit unit 504 is used to receive the initial indicator system plan sent by the processor, and when it is confirmed that the indicators included in the initial indicator system plan are correct, send the audit indicator system to the processor, so that the processor can determine the final indicators included in the audit indicator system, the definition of each of the final indicators, the calculation method of each of the final indicators, and the regulatory name of each of the final indicators.

[0157] It should be understood that the units included in the above-mentioned device for constructing a compliance indicator system based on a large language model are only logical divisions based on the functions implemented by the device. In practical applications, the above-mentioned units can be superimposed or split. In addition, the functions implemented by the device for constructing a compliance indicator system based on a large language model provided in this embodiment correspond one-to-one to the method for constructing a compliance indicator system based on a large language model provided in the above-mentioned embodiment. The more detailed processing flow implemented by the device has been described in detail in the above-mentioned method embodiment 1, and will not be described in detail here.

[0158] Another embodiment of the present invention also provides a computer device, which includes: a processor and a scenario database; the scenario database is used to store computer program code, and the computer program code includes computer instructions; when the processor executes the computer instructions, the electronic device executes each step of the method for constructing a compliance indicator system based on a large language model in the method flow shown in the above method embodiment.

[0159] Another embodiment of the present invention further provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed on a computer device, the computer device executes each step of the method flow shown in the above method embodiment for constructing a compliance indicator system based on a large language model.

[0160] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0161] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for constructing a compliance indicator system based on a large language model, characterized in that: include: The processor receives an instruction for constructing an indicator system sent by a client, analyzes the multi-source data included in the instruction for constructing an indicator system based on a large language model, and obtains the inspection scenarios included in the multi-source data, the regulatory name, scenario type, and scenario name of each inspection scenario, wherein the multi-source data is a regulatory document based on which the indicator system is constructed; The processor receives the scenario classification concept instruction sent by the client, performs cluster analysis on the scenario classification concept instruction and the inspection scenario based on the large language model, obtains an initial scenario classification framework, and sends the initial scenario classification framework to the client, so that the client confirms whether the initial scenario classification framework matches the indicator system concept intention and indicator system restriction conditions included in the indicator system construction instruction; When the processor receives the matching instruction sent by the client, the processor obtains the initial indicator system plan through the large language model according to the indicator system conception intention, the indicator system restriction conditions included in the indicator system construction instruction, the multi-source data and the initial scene classification framework, and sends the initial indicator system plan to the client, so that the client confirms whether the indicators included in the initial indicator system plan are correct; When the processor receives the audit indicator system sent by the client, it determines the final indicators included in the audit indicator system, the definition of each final indicator, the calculation method of each final indicator, and the regulatory name of each final indicator, and sends the audit indicator system to the scenario database.

2. The method according to claim 1, characterized in that The large language model analyzes the multi-source data included in the instruction for building the indicator system, specifically including: The processor identifies the regulatory documents included in the multi-source data, and confirms the compliance check scenario involved in each regulation included in the regulatory documents; Confirm the scenario type, scenario name, and regulatory name and specific clauses corresponding to each inspection scenario.

3. The method according to claim 1, characterized in that The cluster analysis of the scene classification conception instruction and the inspection scene based on the large language model to obtain an initial scene classification framework specifically includes: The processor parses the scene classification conception instruction to obtain at least a desired scene classification method, converts the text of the inspection scene into a vector representation in a high-dimensional vector space, and in the vector space, sequentially determines the similarity of scene types included in the inspection scene according to the similarity between the vectors; Classifying the multiple scene types whose scene type similarity is greater than a first threshold into the same category, and performing hierarchical division according to the scene type similarities of the multiple scene types included in the same category to obtain initial hierarchical clustering; According to the scene classification method, the initial hierarchical clustering is adjusted, and semantic labels are determined for the category assignments included in the updated hierarchical clustering to obtain the initial scene classification framework; the semantic labels are used to summarize the core features of the inspection scenes included in the same category.

4. The method according to claim 1, characterized in that The initial indicator system planning is obtained through a large language model according to the indicator system conception intention, the indicator system restriction conditions included in the indicator system construction instruction, the multi-source data and the initial scene classification framework, specifically including: Determine the characteristics of each type of scene included in the initial scene classification framework, extract the corresponding key evaluation factors of each type of scene, and design indicators based on the key evaluation factors; Determine the calculation method of the indicator according to the conception intention of the indicator system and the restriction conditions of the indicator system; Determine the data source of the indicator according to the multi-source data, and establish the hierarchical relationship and association relationship between the indicators; The initial indicator system planning including the definition of indicators, the calculation method of indicators and the data source of indicators is formed according to the hierarchical relationship and the association relationship.

5. The method according to claim 1, characterized in that The inspection scenarios included in the obtained multi-source data, after the regulatory name, scenario type and scenario name of each inspection scenario, also include: The inspection scenarios included in the multi-source data, the regulatory name of each inspection scenario, the scene type of each inspection scenario and the scene name of each inspection scenario are sent to a scenario database, so that the scenario database stores the inspection scenarios, the regulatory name of each inspection scenario, the scene type of each inspection scenario and the scene name of each inspection scenario.

6. A method for constructing a compliance indicator system based on a large language model, characterized in that: include: Sending an indicator system construction instruction to the processor, so that the processor analyzes the multi-source data included in the indicator system construction instruction to obtain the inspection scenarios included in the multi-source data, and the regulatory name, scenario type and scenario name of each inspection scenario; Sending a scene classification conception instruction to a processor, so that the processor performs cluster analysis on the scene classification conception instruction and the inspection scene and obtains an initial scene classification framework; The initial scene classification framework sent by the processor is received, and when it is confirmed that the initial scene classification framework matches the indicator system conception intention and the indicator system restriction condition included in the indicator system construction instruction, a matching instruction is sent to the processor, so that the processor obtains an initial indicator system plan through a large language model according to the indicator system conception intention, the indicator system restriction condition included in the indicator system construction instruction, the multi-source data and the initial scene classification framework; The initial indicator system plan sent by the processor is received, and when it is confirmed that the indicators included in the initial indicator system plan are correct, the audit indicator system is sent to the processor, so that the processor can determine the final indicators included in the audit indicator system, the definition of each of the final indicators, the calculation method of each of the final indicators, and the regulatory name of each of the final indicators.

7. A device for constructing a compliance indicator system based on a large language model, characterized in that: include: An analysis unit, configured to receive an instruction for constructing an indicator system sent by a client, analyze the multi-source data included in the instruction for constructing an indicator system based on a large language model, and obtain the inspection scenarios included in the multi-source data, the regulatory name, scenario type, and scenario name of each inspection scenario, wherein the multi-source data is a regulatory document based on which the indicator system is constructed; A first obtaining unit is configured to receive a scenario classification concept instruction sent by a client, perform cluster analysis on the scenario classification concept instruction and the inspection scenario based on a large language model, obtain an initial scenario classification framework, and send the initial scenario classification framework to the client, so that the client can confirm whether the initial scenario classification framework matches the indicator system concept intention and indicator system restriction conditions included in the indicator system construction instruction; A second obtaining unit is configured to obtain an initial indicator system plan through a large language model according to the indicator system conception intention, the indicator system restriction conditions included in the indicator system construction instruction, the multi-source data and the initial scene classification framework when the processor receives the matching instruction sent by the client, and send the initial indicator system plan to the client to enable the client to confirm whether the indicators included in the initial indicator system plan are correct; A determination unit is used to determine the final indicators included in the audit indicator system, the definition of each final indicator, the calculation method of each final indicator, and the regulatory name of each final indicator when the processor receives the audit indicator system sent by the client, and send the audit indicator system to the scenario database.

8. A device for constructing a compliance indicator system based on a large language model, characterized in that: include: A first sending unit is used to send an indicator system construction instruction to the processor, so that the processor analyzes the multi-source data included in the indicator system construction instruction to obtain the inspection scenarios included in the multi-source data, and the regulatory name, scenario type and scenario name of each inspection scenario; A second sending unit is used to send a scene classification conception instruction to the processor, so that the processor performs cluster analysis on the scene classification conception instruction and the inspection scene and obtains an initial scene classification framework; A judgment unit, configured to receive the initial scene classification framework sent by the processor, and when confirming that the initial scene classification framework matches the indicator system conception intention and indicator system restriction conditions included in the indicator system construction instruction, send a matching instruction to the processor, so that the processor obtains an initial indicator system plan through a large language model according to the indicator system conception intention, the indicator system restriction conditions included in the indicator system construction instruction, the multi-source data and the initial scene classification framework; The audit unit is used to receive the initial indicator system plan sent by the processor, and when it is confirmed that the indicators included in the initial indicator system plan are correct, send the audit indicator system to the processor, so that the processor can determine the final indicators included in the audit indicator system, the definition of each final indicator, the calculation method of each final indicator, and the regulatory name of each final indicator.

9. A computer device, characterized in that: The computer device includes a scenario database and a processor, wherein the scenario database stores a computer program, and when the computer program is executed by the processor, the processor executes the method for constructing a compliance indicator system based on a large language model as described in any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the processor executes the method for constructing a compliance indicator system based on a large language model as described in any one of claims 1 to 5.

Images