A method, device, equipment and medium for compliance detection of data interaction of a large language model

Abstract

The application provides a compliance detection method, device and equipment for large language model data interaction and a medium, which comprises at least a user interaction detection process, that is, compliance detection is performed on the user request content itself to determine whether there is illegal information, if not, the relevant data stream requested by the user is input to the large language model for processing to obtain a processing result returned by the large language model; the processing result is subjected to compliance detection to determine whether there is illegal information, if yes, the illegal information is filtered and then the processing result is output; the compliance detection comprises sensitive word checking and sensitive information checking. The application at least performs compliance detection on data in the user request interaction stage, thereby realizing the standardized use of the large language model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of large language model data interaction, and in particular to a compliance detection method and device for large language model data interaction, equipment and medium. BACKGROUND

[0002] With the commercialization of large language models, in order to adapt to the regulatory requirements of relevant laws and regulations, follow relevant content review and filtering, data security and personal information protection, and other relevant laws and regulations, it is necessary to effectively and normatively detect products that apply large language models. However, there is currently no relevant method or product. SUMMARY

[0003] The technical problem to be solved by the present application is to provide a compliance detection method and device for large language model data interaction, equipment and medium, which at least detects the data in the user request interaction stage, thereby realizing the normative use of large language models.

[0004] In a first aspect, the present application provides a compliance detection method for large language model data interaction, including a user interaction detection process, the user interaction detection process comprising:

[0005] S1, detecting the compliance of the user request content itself, determining whether there is illegal information, if yes, directly returning the detection result and ending the interaction, if not, proceeding to the next step;

[0006] S3, inputting the relevant data stream of the user request to the large language model for processing, obtaining the processing result returned by the large language model;

[0007] S4, detecting the compliance of the processing result, determining whether there is illegal information, if yes, filtering the illegal information and proceeding to the next step, if not, directly proceeding to the next step;

[0008] S5, outputting the processing result;

[0009] Wherein, the compliance detection includes sensitive word checking and sensitive information checking that do not comply with laws and regulations.

[0010] In a second aspect, the present application provides a compliance detection device for large language model data interaction, including a user interaction detection module, the user interaction detection module being used for executing the following steps:

[0011] S1, detecting the compliance of the user request content itself, determining whether there is illegal information, if yes, directly returning the detection result and ending the interaction, if not, proceeding to the next step;

[0012] S3, inputting the relevant data stream of the user request to the large language model for processing, obtaining the processing result returned by the large language model;

[0013] S4, performing compliance detection on the processing result to determine whether there is illegal information, if yes, filtering the illegal information and then proceeding to the next step, if no, directly proceeding to the next step;

[0014] S5, outputting the processing result;

[0015] The compliance detection includes sensitive word checking and sensitive information checking which do not conform to laws and regulations.

[0016] In a third aspect, the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method of the first aspect when executing the program.

[0017] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, wherein the program is executable on a processor to implement the method of the first aspect.

[0018] The technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: the embodiments of the present application perform compliance detection on the data interaction process (including internal interaction process and external interaction process) of the existing large language model, so that the use of the large language model can comply with the relevant laws and regulations, and make up for the blank of the current large language model without relevant compliance detection. In addition to the compliance detection of relevant laws and regulations, the embodiments of the present application can also perform relevant security detection, such as user access permission checking, usage quota checking, or content authenticity checking, etc., so as to ensure the security of the use of the large language model. The embodiments of the present application also pair and store the user request, the request processing result and / or the supplementary result, so as to retain the interaction history record, for the convenience of security compliance audit in the future.

[0019] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the embodiments of the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, characteristics and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application are described below. BRIEF DESCRIPTION OF DRAWINGS

[0020] The present application will be further described below with reference to the accompanying drawings and embodiments.

[0021] Figure 1 It is a schematic diagram of the framework of the system of the present application;

[0022] Figure 2 It is a functional schematic diagram of the security compliance detection module of the present application;

[0023] Figure 3A flowchart of the method in Embodiment One of the present application;

[0024] Figure 4 A structural schematic diagram of the device in Embodiment Two of the present application;

[0025] Figure 5 A structural schematic diagram of the electronic device in Embodiment Three of the present application;

[0026] Figure 6 A structural schematic diagram of the medium in Embodiment Four of the present application. DETAILED DESCRIPTION

[0027] The embodiments of the present application provide a compliance detection method, device, equipment and medium for data interaction of a large language model, which at least performs compliance detection on data in a user request interaction stage, so as to realize standard use of the large language model.

[0028] The technical solutions in the embodiments of the present application have the following general idea: performing compliance detection on the data interaction process (including internal interaction process and external interaction process) of an existing large language model, so that the use of the large language model can comply with the provisions of relevant laws and regulations, and make up for the current blank of no relevant compliance detection of the large language model. In addition to compliance detection of relevant laws and regulations, the embodiments of the present application can also simultaneously perform relevant safety detection, such as user access permission check, usage quota check or content authenticity check, so as to ensure the safety of use of the large language model. The embodiments of the present application also store the pairing of the user request, the request processing result and / or the supplementary result, so as to retain the interaction history record, for the convenience of safety compliance audit in the future.

[0029] Before introducing the specific embodiments, the system framework corresponding to the method of the embodiments of the present application is introduced first, as shown in Figure 1 and Figure 2 The system is roughly divided into three parts:

[0030] The large language model is a machine learning technology used to study the probability distribution of natural language data and utilize these distributions to complete language-related tasks, such as text classification, natural language understanding, machine translation, etc. The interaction process is to receive a request of a user, perform relevant processing on the content of the request, and return a processing result.

[0031] The data storage module is also needed to build a private knowledge base for a large language model product with a private knowledge base; the private knowledge base can supplement information for the request of the user.

[0032] A security compliance detection module is configured to perform security compliance detection (mainly including sensitive word checking that does not comply with laws and regulations and sensitive information checking) and security detection (mainly including user access permission checking, usage quota checking, or fake content, etc.) on the large language model data interaction process, so as to ensure that the use of the large language model complies with the relevant laws and regulations and is safe. Specifically, the security compliance detection module includes:

[0033] Sensitive word checking, that is, sensitive word checking that does not comply with laws and regulations;

[0034] Sensitive information checking, that is, checking related to enterprise secrets, personal privacy information, etc.

[0035] User access permission checking, that is, managing file permissions;

[0036] Usage quota checking, that is, controlling interaction costs to achieve financial audit purposes;

[0037] Interface authentication, avoiding being attacked or maliciously attacked when calling external interfaces;

[0038] Trusted source checking, confirming the trusted source of external data (similar to a whitelist) when accessing external data to avoid pollution by returned external data;

[0039] Authenticity checking, checking whether the returned results are real and not fabricated, but in creative generation interactions, the authenticity detection weight can be reduced to ensure smooth generation of creativity;

[0040] Audit business, archiving historical logs for future security audits and problem tracing.

[0041] Embodiment One

[0042] As shown in Figure 3 , the embodiment provides a compliance detection method for large language model data interaction, including a user interaction detection process, the user interaction detection process including:

[0043] S1, performing compliance detection on the user request content itself, determining whether there is illegal information, if so, directly returning the detection result and ending the interaction, if not, proceeding to the next step;

[0044] S3, inputting the related data stream of the user request to the large language model for processing to obtain a processing result returned by the large language model;

[0045] S4, performing compliance detection on the processing result, determining whether there is illegal information, if so, filtering the illegal information and proceeding to the next step, if not, directly proceeding to the next step;

[0046] S5, outputting the processing result;

[0047] The compliance detection includes sensitive word checking and sensitive information checking that do not comply with laws and regulations, so as to avoid transmission of sensitive words and leakage of sensitive information.

[0048] Further, as a more optimal or more specific implementation manner of the embodiment, the user interaction detection process further includes:

[0049] S2, judging whether there is a part that needs to access the knowledge base and rely on the knowledge base as supplementary information in the user request content, if yes, performing the compliance check and the user access permission check on the supplementary result returned by the knowledge base, judging whether the supplementary result contains the illegal information or the information of unauthorized access, if yes, filtering the illegal information or the information of unauthorized access and returning, and entering S3, if no, entering S3. Thus, the safety of sensitive information leakage or unauthorized access caused by the supplementary result returned by the knowledge base is avoided.

[0050] In S1, the user request content itself is also subjected to the user access permission check and the usage quota check, to judge whether the user request content contains unauthorized access or usage over-limit; if yes, the detection result is directly returned and the interaction is ended, if no, the next step is performed;

[0051] In S2, the supplementary result is also subjected to the user access permission check, to judge whether the supplementary result contains unauthorized access; if yes, the content of unauthorized access is filtered and returned, and S3 is entered, if no, S3 is entered;

[0052] In S3, interface authentication is performed on the call initiated by the large language model, and if the authentication is passed, the next step is performed, if not, the authentication result is returned;

[0053] In S4, the processing result is also subjected to the user access permission check and the content authenticity check, to judge whether the processing result contains unauthorized access or false content; if yes, the processing result is filtered and the next step is performed, if no, the next step is directly performed;

[0054] S5 further includes: pairing the user request content with the processing result and / or the supplementary result and storing.

[0055] Further, as a more optimal or more specific implementation manner of the embodiment, the method of the embodiment further includes at least one of the following processes:

[0056] Private knowledge base construction or update detection process: During the construction or update of the private knowledge base, the incoming documents are subjected to compliance checks and user access permission checks to determine whether the documents contain illegal information or unauthorized access. If not, the documents, document content, and related parameters are stored in the knowledge base. If so, a risk warning is issued, user permissions are confirmed, and after user authorization, relevant processing logs are recorded, marking the risk level of the documents, document content, and user level information. Finally, the documents, document content, and related parameters are stored in the knowledge base.

[0057] External data access interaction detection process: When the large language model accesses an external data source, the external data returned by the interaction is subjected to compliance detection, trusted source detection, and content authenticity detection to determine whether the external data contains illegal information, untrusted sources, or false content. If so, the external data is filtered to ensure that the returned external data does not contain sensitive information and that the data source is credible and the data is authentic and reliable.

[0058] Functional module call interaction detection process: When the large language model calls a functional module, the functional data returned by the functional module interaction undergoes compliance detection, trusted source detection, and content authenticity check to determine whether the functional data contains illegal information, untrusted sources, or false content. If so, the functional data is filtered. This ensures that the returned functional data does not contain sensitive information and that the data source is trustworthy and the data is authentic and reliable.

[0059] Based on the same inventive concept, this application also provides an apparatus corresponding to the method in Embodiment 1, as detailed in Embodiment 2.

[0060] Example 2

[0061] like Figure 4 As shown, this embodiment provides a compliance detection device for large language model data interaction, including a user interaction detection module, which is used to perform the following steps:

[0062] S1. Perform compliance checks on the user's requested content to determine if there is any illegal information. If so, return the detection result directly and end the interaction. If not, proceed to the next step.

[0063] S3. Input the relevant data stream of the user request into the large language model for processing, and obtain the processing result returned by the large language model;

[0064] S4. Perform a compliance check on the processing result to determine whether there is any violation information. If so, filter the violation information and proceed to the next step. If not, proceed directly to the next step.

[0065] S5, output the processing result;

[0066] The compliance detection includes sensitive word checking and sensitive information checking that do not comply with laws and regulations.

[0067] Further, as a more optimal or more specific implementation manner of the embodiment, the user interaction detection module is further used to perform the following steps after performing the S1:

[0068] S2, determine whether there is a part in the user request content that needs to access the knowledge base and rely on the knowledge base as supplementary information, if yes, perform the compliance check and user access permission check on the supplementary result returned by the knowledge base, determine whether the supplementary result contains the illegal information or information of unauthorized access, if yes, return the illegal information or information of unauthorized access after filtering, and enter the S3, if not, enter the S3.

[0069] In the S1, the user request content itself is also subjected to the user access permission check and usage quota check, to determine whether the user request content contains unauthorized access or usage exceeding the limit; if yes, directly return the detection result and end the interaction, if not, proceed to the next step;

[0070] In the S2, the supplementary result is also subjected to the user access permission check, to determine whether the supplementary result contains unauthorized access; if yes, return the content of unauthorized access after filtering, and enter the S3, if not, enter the S3;

[0071] In the S3, interface authentication is also performed for the call initiated by the large language model, if the authentication is passed, proceed to the next step, if not, return the authentication result;

[0072] In the S4, the processing result is also subjected to the user access permission check and content authenticity check; determine whether the processing result contains unauthorized access or false content, if yes, filter the processing result and proceed to the next step, if not, directly proceed to the next step;

[0073] The S5 also includes: pairing the user request content with the processing result and / or the supplementary result and storing them.

[0074] Further, as a more optimal or more specific implementation manner of the embodiment, the device of the embodiment also includes at least one of the following modules:

[0075] The private knowledge base construction or update detection module is used for performing the compliance detection and user access permission check on the incoming document when the private knowledge base is constructed or updated, judging whether the document has illegal information or unauthorized access, if not, storing the document, document content and related parameters into the knowledge base, if yes, prompting the document to store related risks, confirming user permissions, recording related processing logs after user authorization, marking document, document content risk level and user level information, and then storing the document, document content and related parameters into the knowledge base.

[0076] The external data access interaction detection module is used for performing the compliance detection, trusted source detection and content authenticity detection on the external data returned by the interaction when the large language model accesses the external data source, judging whether the external data has illegal information, untrusted source and false content, if yes, filtering the external data.

[0077] The function module calling interaction detection module is used for performing the compliance detection, trusted source detection and content authenticity check on the function data returned by the function module interaction when the large language model calls the function module, judging whether the function data has illegal information, untrusted source and false content, if yes, filtering the function data.

[0078] Since the device introduced in the second embodiment of the application is the device used for implementing the method of the first embodiment of the application, the specific structure and deformation of the device can be understood by those skilled in the art based on the method introduced in the first embodiment of the application, and therefore will not be described here again.

[0079] Based on the same inventive concept, the present application provides electronic device embodiments corresponding to the first embodiment, which are described in detail in Embodiment Three.

[0080] Embodiment Three

[0081] The present embodiment provides an electronic device, as shown in the accompanying drawings, comprising a memory, a processor and a computer program stored in the memory and executable on the processor. Figure 5 The processor can implement any of the embodiments of the first embodiment when executing the computer program.

[0082] Since the electronic device introduced in this embodiment is the device used to implement the method in Embodiment One of the present application, the specific implementation of the electronic device of this embodiment and its various forms can be understood by those skilled in the art based on the method introduced in Embodiment One of the present application, so the electronic device how to implement the method in the present embodiment will not be introduced in detail. As long as the device used to implement the method in the present embodiment is implemented by those skilled in the art, it belongs to the scope of the present application.

[0083] Based on the same inventive concept, the present application provides a storage medium corresponding to Embodiment One, which is described in detail in Embodiment Four.

[0084] Embodiment Four

[0085] The present embodiment provides a computer-readable storage medium, such as Figure 6 As shown in the figure, the computer program stored thereon can implement any of the embodiments in Embodiment One when executed by a processor.

[0086] The technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: the method, device, system, equipment and medium provided in the embodiments of the present application,

[0087] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, system, or computer program product. Therefore, the present application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0088] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The device that implements the functions specified in one or more flows and / or blocks.

[0089] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 The flow or flows and / or blocks Figure 1 The flow or flows and / or blocks

[0090] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 The flow or flows and / or blocks Figure 1 The flow or flows and / or blocks

[0091] While the above describes a particular embodiment of the application, those skilled in the art will understand that the specific embodiments described are only illustrative of the present application and are not intended to limit the scope of the present application, which is defined by the following claims.

Claims

1. A compliance detection method for large language model data interaction, characterized in that: This includes a user interaction detection process, which includes: S1. Perform compliance checks on the user's requested content to determine if there is any illegal information. If so, return the detection result directly and end the interaction. If not, proceed to the next step. S2. Determine whether there is a part in the user's request content that requires access to the knowledge base or relies on the knowledge base as supplementary information. If so, when the knowledge base returns the supplementary result, perform the compliance detection and user access permission check on the supplementary result, and determine whether there is any violation information or unauthorized access information in the supplementary result. If so, filter the violation information or unauthorized access information and return it, and proceed to S3. If not, proceed to S3. S3. Input the relevant data stream of the user request into the large language model for processing, and obtain the processing result returned by the large language model; S4. Perform a compliance check on the processing result to determine whether there is any violation information. If so, filter the violation information and proceed to the next step. If not, proceed directly to the next step. S5. Output the processing result; The compliance testing includes checking for sensitive words and phrases that do not comply with laws and regulations, and checking for sensitive information. S1 also includes checking the user's access permissions and usage limits for the user's requested content to determine whether the user's requested content has unauthorized access or excessive usage. If so, the detection result is returned directly and the interaction ends; otherwise, the next step is performed. In step S2, the supplementary result is further checked for user access permissions to determine whether there is any unauthorized access. If so, the unauthorized access content is filtered out and returned, and the process proceeds to step S3. If not, the process proceeds to step S3. In S3, interface authentication is also performed on the calls initiated by the large language model. If the authentication is successful, the next step is performed; if the authentication fails, the authentication result is returned. In step S4, the processing result is further checked for user access permissions and content authenticity; it is determined whether the processing result contains unauthorized access or false content. If so, the processing result is filtered and the next step is performed; otherwise, the next step is performed directly. S5 further includes: pairing the user request content with the processing result and / or the supplementary result and then storing it.

2. The compliance detection method for large language model data interaction according to claim 1, characterized in that: It also includes at least one of the following processes: Private knowledge base construction or update detection process: When building or updating the private knowledge base, the incoming documents are subjected to compliance checks and user access permission checks to determine whether the documents contain illegal information or have unauthorized access. If not, the documents, document content and related parameters are stored in the knowledge base. If so, the documents are prompted that there are related risks, user permissions are confirmed, and after user authorization, relevant processing logs are recorded, the risk level of the documents and document content and user level information are marked, and then the documents, document content and related parameters are stored in the knowledge base. External data access interaction detection process: When the large language model accesses an external data source, the external data returned by the interaction is subjected to compliance detection, trusted source detection and content authenticity detection to determine whether the external data contains illegal information, untrusted sources and false content. If so, the external data is filtered. Functional module call interaction detection process: When the large language model calls a functional module, the functional data returned by the functional module interaction is subjected to compliance detection, trust source detection, and content authenticity check to determine whether the functional data contains illegal information, untrusted sources, or false content. If so, the functional data is filtered.

3. A compliance detection device for large language model data interaction, characterized in that: Includes a user interaction detection module, which is used to perform the following steps: S1. Perform compliance checks on the user's requested content to determine if there is any illegal information. If so, return the detection result directly and end the interaction. If not, proceed to the next step. S2. Determine whether there is a part in the user's request content that requires access to the knowledge base or relies on the knowledge base as supplementary information. If so, when the knowledge base returns the supplementary result, perform the compliance detection and user access permission check on the supplementary result, and determine whether there is any violation information or unauthorized access information in the supplementary result. If so, filter the violation information or unauthorized access information and return it, and proceed to S3. If not, proceed to S3. S3. Input the relevant data stream of the user request into the large language model for processing, and obtain the processing result returned by the large language model; S4. Perform a compliance check on the processing result to determine whether there is any violation information. If so, filter the violation information and proceed to the next step. If not, proceed directly to the next step. S5. Output the processing result; The compliance detection includes checking for sensitive words and phrases that do not comply with laws and regulations, and checking for sensitive information; and the user interaction detection module is also used for: S1 also includes checking the user's access permissions and usage limits for the user's requested content to determine whether the user's requested content has unauthorized access or excessive usage. If so, the detection result is returned directly and the interaction ends; otherwise, the next step is performed. In step S2, the supplementary result is further checked for user access permissions to determine whether there is any unauthorized access. If so, the unauthorized access content is filtered out and returned, and the process proceeds to step S3. If not, the process proceeds to step S3. In S3, interface authentication is also performed on the calls initiated by the large language model. If the authentication is successful, the next step is performed; if the authentication fails, the authentication result is returned. In step S4, the processing result is further checked for user access permissions and content authenticity; it is determined whether the processing result contains unauthorized access or false content. If so, the processing result is filtered and the next step is performed; otherwise, the next step is performed directly. S5 further includes: pairing the user request content with the processing result and / or the supplementary result and then storing it.

4. The compliance detection device for large language model data interaction according to claim 3, characterized in that: It also includes at least one of the following modules: Private Knowledge Base Construction or Update Detection Module: When building or updating a private knowledge base, this module performs compliance checks and user access permission checks on the incoming documents to determine whether the documents contain illegal information or unauthorized access. If not, the document, document content, and related parameters are stored in the knowledge base. If so, the module prompts that the document has related risks, confirms user permissions, records relevant processing logs after user authorization, marks the risk level of the document, document content, and user level information, and then stores the document, document content, and related parameters in the knowledge base. External data access interaction detection module: When the large language model accesses external data sources, it performs compliance detection, trusted source detection, and content authenticity detection on the external data returned by the interaction, and determines whether the external data contains illegal information, untrusted sources, or false content. If so, it filters the external data. Functional module call interaction detection module: When a large language model calls a functional module, it performs compliance detection, trust source detection, and content authenticity check on the functional data returned by the functional module interaction to determine whether the functional data contains illegal information, untrusted sources, or false content. If so, the functional data is filtered.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in claim 1 or 2.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in claim 1 or 2.

Images