Large model security management method based on organization isolation and authority control

Abstract

The invention is suitable for the technical field of human-computer interaction, and provides a large model security management method based on organization isolation and authority control. The method comprises the following steps: acquiring a user question and a user ID sent by a service system; user role information is determined according to the user ID, a definitive prompt word is generated based on the user role information, and the definitive prompt word is used for limiting a response information range corresponding to the user question; retrieving a corresponding response information fragment from a preset knowledge base according to the user question and the restrictive cue word; generating response information according to the response information fragment; and detecting whether the response information contains the unauthorized content or not through a rule engine, and sending the response information to the service system when it is not detected that the response information contains the unauthorized content. According to the application, the response information can be strictly matched with the user permission and the service scene, redundant interaction is reduced through structuring personnel information, and sensitive information leakage or unauthorized access is avoided.

Description

Technical Field

[0001] This application relates to the field of human-computer interaction technology, specifically to a large-scale security management method based on organizational isolation and access control. Background Technology

[0002] In recent years, with the rapid development of artificial intelligence (AI) technology, its applications in various fields have become increasingly widespread, and more and more intelligent products based on human-computer interaction technology have emerged. Large-scale models have achieved remarkable results in various question-answering tasks, accurately understanding and responding to human instructions, enabling them to correctly understand and answer complex questions, and have been widely applied in various intelligent question-answering scenarios.

[0003] When enterprises use large models to handle internal business problems, they usually face the following challenges: (1) Information overreach: employees of different job levels and departments may obtain sensitive information beyond their authority; (2) Answer generalization: the answers returned by the large model lack specificity and cannot be combined with the user's specific organizational background; (3) Inefficiency: the context information such as organization and role needs to be manually entered, and the interaction process is cumbersome. Summary of the Invention

[0004] In view of this, embodiments of this application provide a large-scale security management method based on organizational isolation and access control, which enables response information to strictly match user permissions and business scenarios, reduces redundant interactions through structured personnel information, and avoids leakage of sensitive information or unauthorized access.

[0005] To achieve the above objectives, this application adopts the following technical solution:

[0006] This application provides a large-scale security management method based on organizational isolation and access control, including:

[0007] Retrieve user questions and user IDs sent by the business system;

[0008] User role information is determined based on the user ID, and limiting prompt words are generated based on the user role information. The limiting prompt words are used to limit the range of response information corresponding to the user's question.

[0009] Based on the user's question and the limiting prompts, retrieve the corresponding response information fragments from the preset knowledge base;

[0010] Generate response information based on the response information fragment;

[0011] The rule engine detects whether the response information contains unauthorized content. If no unauthorized content is detected in the response information, the response information is sent to the business system.

[0012] In some embodiments, obtaining the user question and user ID sent by the business system includes:

[0013] The business system receives account information and user questions input by the user, and determines the user ID based on the account information;

[0014] Send the user's question and the user's ID to the server.

[0015] In some embodiments, determining user role information based on the user ID and generating limiting prompts based on the user role information includes:

[0016] The user's organization, department, and position are determined based on the user ID and a preset employee information database. The preset employee information database stores multiple employee information entries, each of which includes the employee ID, the organization, department, and position.

[0017] Generate qualifying prompts based on the user's organization, department, and position, wherein the qualifying prompts include the user's organization, department, and position so that the response information matches the user's organization, department, and position.

[0018] In some embodiments, the method includes:

[0019] Obtain corporate policies and regulations;

[0020] The enterprise policy documents are parsed using an NLP model to obtain the metadata text of the enterprise policy documents. The metadata includes at least one of the following: title, scope of application, and authority clauses.

[0021] The metadata text is converted into a vector and stored in a preset knowledge base.

[0022] In some embodiments, retrieving corresponding response information fragments from a preset knowledge base based on the user question and the limiting prompts includes:

[0023] Transform the user question into a first vector;

[0024] The target knowledge information is determined by matching the first vector from the preset knowledge base, and each piece of knowledge information in the preset knowledge base is set with an access permission label;

[0025] Based on the limiting prompt and the permission tag, determine the knowledge information corresponding to the limiting prompt from the target knowledge information.

[0026] In some embodiments, the step of matching and determining target knowledge information from the preset knowledge base based on the first vector includes:

[0027] Calculate the distance between the vector corresponding to each piece of knowledge information in the preset knowledge base and the first vector;

[0028] Knowledge information whose distance is less than a threshold is identified as the target knowledge information.

[0029] In some embodiments, determining the knowledge information corresponding to the limiting prompt word from the target knowledge information based on the limiting prompt word and the permission tag includes:

[0030] Match the organization, department, and position in the limiting prompt with the permission tags of the target knowledge information;

[0031] The target knowledge information that is successfully matched is used as the response information fragment.

[0032] In some embodiments, generating response information based on the response information fragment includes:

[0033] The keywords of the response information fragments are determined. These keywords are either the title or scope of application of the file corresponding to the response information fragment, or words in the response information fragment. The various response information fragments are distinguished by these keywords.

[0034] The keywords are sorted according to a preset rule; wherein the preset rule is the priority between keywords, and the response information fragments corresponding to keywords with higher priority are arranged first, and the response information fragments corresponding to keywords with lower priority are arranged last.

[0035] The response information is synthesized from the various response information fragments based on the order of the keywords.

[0036] In some embodiments, the method further includes:

[0037] When unauthorized content is detected in the response information, the field corresponding to the unauthorized content in the response information is hidden, and then the response information is sent to the business system.

[0038] In some embodiments, sending the response information to the business system includes:

[0039] Add applicable boundaries to the response information and send the response information to the business system.

[0040] The beneficial effects of the embodiments of this application compared with the prior art include:

[0041] In this embodiment, the user question and user ID sent by the business system are first obtained. Then, the user role information is determined based on the user ID, and a limiting prompt word is generated based on the user role information. Next, the corresponding response information fragments are retrieved from the preset knowledge base based on the user question and the limiting prompt word. The response information fragments are then combined to form response information. Finally, the rule engine detects whether the response information contains unauthorized content. If no unauthorized content is detected in the response information, the response information is sent to the business system, thereby realizing the automatic generation of response information for the user's input question.

[0042] This application embodiment automatically transmits the user's organizational structure, position, and role information through the user ID, constructs a dynamic personnel system in the large model, and automatically attaches permission and scope limitation conditions to the user's questions in combination with preset prompt word rules, so that the response information strictly matches the user's permissions and business scenarios. By structuring personnel information, redundant interactions are reduced, and sensitive information leakage or unauthorized access is avoided. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a schematic diagram of an application scenario provided in the embodiments of this application;

[0045] Figure 2 This is a flowchart illustrating the large-scale security management method based on organizational isolation and access control provided in this application embodiment.

[0046] Figure 3 This is a schematic diagram of the server structure provided in an embodiment of this application. Detailed Implementation

[0047] The present application will be described more clearly below with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the function of the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.

[0048] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0049] See Figure 1In this application scenario, a terminal 10 and a server 20 are included. A business system runs on the terminal 10. When a user needs to query relevant provisions in a company's policy documents (e.g., insurance fee rules), the user can log in to the business system through the terminal 10, enter a query question in the business system, and the business system sends the question and user ID to the server 20. After receiving the user's question and user ID from the business system, the server 20 generates a corresponding response based on the question and user ID and sends it to the business system. The user can then view the corresponding response information after entering a question through the business system.

[0050] The following combination Figure 1 The embodiments of this application will be described in detail.

[0051] See Figure 2 The large-scale security management method based on organizational isolation and access control provided in this application embodiment can be applied to the server 20 side, specifically including the following steps:

[0052] Step 101: Obtain the user question and user ID sent by the business system.

[0053] In some embodiments, obtaining the user question and user ID sent by the business system may include: the business system receiving account information and user question input by the user, determining the user ID based on the account information, and sending the user question and user ID to the server.

[0054] In some scenarios, when a user needs to query relevant provisions in a company's policy documents (such as insurance fee rules), the user can log in to the business system and then enter the question they want to ask. After the business system detects that the user has logged in, it can match the user's entered account information with pre-stored user information (which may include a user ID and a user account), and retrieve the user ID from the successfully matched user information.

[0055] Step 102: Determine user role information based on the user ID, and generate limiting prompt words based on the user role information. The limiting prompt words are used to limit the range of response information corresponding to the user's question.

[0056] In some embodiments, step 102 may include: determining the user's organization, department, and position based on the user ID (Identity document) and a preset employee information database, wherein the preset employee information database stores multiple employee information entries, each including the employee ID and the organization, department, and position; generating limiting prompts based on the user's organization, department, and position, wherein the limiting prompts include the user's organization, department, and position so that the response information matches the user's organization, department, and position.

[0057] For example, if a user wants to query the 2023 travel expense standards, and the user ID indicates that the user's organization is the expense auditor of Group A's East China region, then the generated limiting prompt can be "The user is an expense auditor in the East China region, please only return the 2023 travel expense standards for the East China region".

[0058] In some embodiments, the above-described large-scale security management method based on organizational isolation and access control may include: obtaining corporate policy documents; parsing the corporate policy documents using an NLP model to obtain metadata text of the corporate policy documents, wherein the metadata includes at least one of title, scope of application, and access control clauses; and converting the metadata text into a vector and storing it in a preset knowledge base.

[0059] Specifically, it can obtain the company's institutional documents, financial policy documents, etc., and then use an NLP (Natural Language Processing) model to parse the above-mentioned institutional documents, financial policy documents, etc., extract metadata text such as title, scope of application, and authority clauses, and then convert the metadata text into vectors and store them in a preset knowledge base.

[0060] For example, the process of parsing a file using an NLP model can be as follows: First, the file is converted to a plain text file. Then, the plain text file is cleaned to remove headers, footers, special characters, etc. Next, the text is segmented into paragraphs and sentences. Then, headings located in preset positions within the text and formatted according to a preset format (headings are usually at the beginning of the text and have a special format) are extracted. The applicable scope is searched using keywords, and the relevant permissions are determined by searching for chapters or keywords.

[0061] Optionally, the generation rules for the limiting prompt words can be flexibly configured in this application embodiment to adapt to various business scenarios.

[0062] Step 103: Retrieve the corresponding response information fragment from the preset knowledge base based on the user question and the limiting prompt words.

[0063] In some embodiments, step 103 may include: converting the user question into a first vector; matching and determining target knowledge information from the preset knowledge base based on the first vector, wherein each piece of knowledge information in the preset knowledge base is assigned a permission tag; and determining the knowledge information corresponding to the limiting prompt word from the target knowledge information based on the limiting prompt word and the permission tag.

[0064] For example, the step of matching and determining target knowledge information from the preset knowledge base based on the first vector may include: calculating the distance between the vector corresponding to each piece of knowledge information in the preset knowledge base and the first vector; and determining the knowledge information with a distance less than a threshold as the target knowledge information. The distance between the vector corresponding to each piece of knowledge information in the preset knowledge base and the first vector can be a cosine distance or other distances, which are not limited thereto. Alternatively, the distances can be sorted in ascending order to obtain a ranking of the knowledge information in the preset knowledge base; and the top one or more pieces of knowledge information in the ranking can be determined as the target knowledge information.

[0065] Alternatively, the user question can be converted into a first keyword vector and a first semantic vector, and each piece of knowledge information in the preset knowledge base can be converted into a second keyword vector and a second semantic vector; the first distance between the second keyword vector and the first keyword vector of each piece of knowledge information, and the second distance between the second semantic vector and the first semantic vector can be calculated, and the comprehensive distance between the user question and each piece of knowledge information can be determined based on the first distance and the second distance; knowledge information with a comprehensive distance less than a threshold can be determined as target knowledge information.

[0066] For example, determining the knowledge information corresponding to the limiting prompt word from the target knowledge information based on the limiting prompt word and the permission tag may include: matching the organization, department, and position in the limiting prompt word with the permission tag of the target knowledge information; and using the successfully matched target knowledge information as a response information fragment.

[0067] Specifically, the permission tags for knowledge information can represent the relationship between an organization, department, or position and the permissions of that knowledge information. For example, permission tags can be "visible only to employees of Department B" or "visible only to managers." That is, if the permission tag for a piece of knowledge information is "visible only to employees of Department B," then employees outside of Department B do not have permission to view it; if the permission tag for a piece of knowledge information is "visible only to managers," then employees at the manager level and above have permission to view it.

[0068] Step 104: Generate response information based on the response information fragment.

[0069] In some embodiments, step 104 may include: determining the keywords of the response information fragments; sorting the keywords according to a preset rule; and synthesizing the response information fragments into response information based on the sorting of the keywords.

[0070] Specifically, keywords can be information such as the title and scope of the file corresponding to the response information fragment, or words within the response information fragment. These keywords can be used to distinguish between different response information fragments. The aforementioned preset rules can be used to prioritize keywords. For example, by setting priorities for each keyword, response information fragments corresponding to higher-priority keywords are arranged first, and those corresponding to lower-priority keywords are arranged last, thus generating response information. The priority of these keywords can be set according to the user's level of attention, setting the priority of keywords with higher attention to high and the priority of keywords with lower attention to low. The user's level of attention to keywords can be determined by the number or frequency of occurrence of the keywords in previously generated response information. A higher frequency or frequency of occurrence of keywords in previously generated response information indicates a higher level of user attention to the keywords; conversely, a lower frequency or frequency indicates a lower level of user attention to the keywords.

[0071] Step 105: Detect whether the response information contains unauthorized content using the rule engine. If no unauthorized content is detected in the response information, send the response information to the business system.

[0072] In some embodiments, the large-scale security management method based on organizational isolation and access control may further include: when unauthorized content is detected in the response information, hiding the field corresponding to the unauthorized content in the response information, and then sending the response information to the business system.

[0073] For example, if the response information contains unauthorized content, a preset flag (such as *) can be used to replace the field corresponding to the unauthorized content, thereby achieving desensitization of sensitive information.

[0074] Optionally, the permission tags for the aforementioned knowledge information can be dynamically adjusted as needed. By dynamically adjusting the permission tags for knowledge information, the risk of leakage of sensitive information can be greatly reduced.

[0075] In this embodiment of the application, the generated response information is verified by a rule engine, thereby reducing the manual permission verification process and greatly shortening the response time between the user inputting the question and the generation of the response information.

[0076] In some embodiments, sending the response information to the business system may include: adding an applicable boundary to the response information and then sending the response information to the business system.

[0077] For example, the applicable boundaries of the information can be marked in the response information. For instance, the applicable boundary of the information can be "Based on your permissions, the following is data for East China".

[0078] The aforementioned large-scale security management method based on organizational isolation and access control first obtains the user question and user ID sent by the business system. Then, it determines the user role information based on the user ID and generates limiting prompts based on the user role information. Next, it retrieves the corresponding response information fragments from a preset knowledge base based on the user question and limiting prompts. The response information fragments are then combined to form a response information. Finally, a rule engine detects whether the response information contains unauthorized content. If no unauthorized content is detected in the response information, the response information is sent to the business system, thereby realizing the automatic generation of response information for user-input questions.

[0079] This application embodiment automatically transmits the user's organizational structure, position, and role information through the user ID, constructs a dynamic personnel system in the large model, and automatically attaches permission and scope limitation conditions to the user's questions in combination with preset prompt word rules, so that the response information strictly matches the user's permissions and business scenarios. By structuring personnel information, redundant interactions are reduced, and sensitive information leakage or unauthorized access is avoided.

[0080] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0081] Figure 3 This is a schematic diagram of a server provided in an embodiment of the present invention. Figure 3 As shown, the server 20 in this embodiment includes a processor 21 and a memory 22. The memory 22 stores computer programs that can run on the processor 21, such as a large-scale security management program based on organizational isolation and access control. When the processor 21 executes the computer program, it implements the steps in the above-described embodiment of the large-scale security management method based on organizational isolation and access control, for example... Figure 2 Steps 101 to 105 are shown.

[0082] For example, the computer program may be divided into one or more modules / units, which are stored in the memory 22 and executed by the processor 21 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the server 20.

[0083] The server 20 can be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor 21 and a memory 22. Those skilled in the art will understand that... Figure 3 This is merely an example of server 20 and does not constitute a limitation on server 20. It may include more or fewer components than shown, or combine certain components, or different components. For example, the terminal device may also include input / output devices, network access devices, buses, etc.

[0084] The processor 22 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0085] The memory 22 can be an internal storage unit of the server 20, such as a hard drive or memory of the server 20. The memory 22 can also be an external storage device of the server 20, such as a plug-in hard drive, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the server 20. Furthermore, the memory 22 can include both internal storage units and external storage devices of the server 20. The memory 22 is used to store the computer program and other programs and data required by the terminal device. The memory 22 can also be used to temporarily store data that has been output or will be output.

[0086] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A large-scale security management method based on organizational isolation and access control, characterized in that, Applied to the server side, the method includes: Retrieve user questions and user IDs sent by the business system; User role information is determined based on the user ID, and limiting prompt words are generated based on the user role information. The limiting prompt words are used to limit the range of response information corresponding to the user's question. Based on the user's question and the limiting prompts, retrieve the corresponding response information fragments from the preset knowledge base; Generate response information based on the response information fragment; The rule engine detects whether the response information contains unauthorized content. If no unauthorized content is detected in the response information, the response information is sent to the business system.

2. The large-scale security management method based on organizational isolation and access control according to claim 1, characterized in that, The process of obtaining the user question and user ID sent by the business system includes: The business system receives account information and user questions input by the user, and determines the user ID based on the account information; Send the user's question and the user's ID to the server.

3. The large-scale security management method based on organizational isolation and access control according to claim 1, characterized in that, The step of determining user role information based on the user ID and generating limiting prompts based on the user role information includes: The user's organization, department, and position are determined based on the user ID and a preset employee information database. The preset employee information database stores multiple employee information entries, each of which includes the employee ID, the organization, department, and position. Generate qualifying prompts based on the user's organization, department, and position, wherein the qualifying prompts include the user's organization, department, and position so that the response information matches the user's organization, department, and position.

4. The large-scale security management method based on organizational isolation and access control according to claim 3, characterized in that, The method includes: Obtain corporate policies and regulations; The enterprise policy documents are parsed using an NLP model to obtain the metadata text of the enterprise policy documents. The metadata includes at least one of the following: title, scope of application, and authority clauses. The metadata text is converted into a vector and stored in a preset knowledge base.

5. The large-scale security management method based on organizational isolation and access control according to claim 4, characterized in that, The step of retrieving corresponding response information fragments from a preset knowledge base based on the user question and the limiting prompts includes: Transform the user question into a first vector; The target knowledge information is determined by matching the first vector from the preset knowledge base, and each piece of knowledge information in the preset knowledge base is set with an access permission label; Based on the limiting prompt and the permission tag, determine the knowledge information corresponding to the limiting prompt from the target knowledge information.

6. The large-scale security management method based on organizational isolation and access control according to claim 5, characterized in that, The step of matching and determining target knowledge information from the preset knowledge base based on the first vector includes: Calculate the distance between the vector corresponding to each piece of knowledge information in the preset knowledge base and the first vector; Knowledge information whose distance is less than a threshold is identified as the target knowledge information.

7. The large-scale security management method based on organizational isolation and access control according to claim 5, characterized in that, The step of determining the knowledge information corresponding to the limiting prompt word from the target knowledge information based on the limiting prompt word and the permission tag includes: Match the organization, department, and position in the limiting prompt with the permission tags of the target knowledge information; The target knowledge information that is successfully matched will be used as the response information fragment.

8. The large-scale security management method based on organizational isolation and access control according to claim 1, characterized in that, The step of generating response information based on the response information fragment includes: The keywords of the response information fragments are determined. These keywords are either the title or scope of application of the file corresponding to the response information fragment, or words in the response information fragment. The various response information fragments are distinguished by these keywords. The keywords are sorted according to a preset rule; wherein the preset rule is the priority between keywords, and the response information fragments corresponding to keywords with higher priority are arranged first, and the response information fragments corresponding to keywords with lower priority are arranged last. The response information is synthesized from the various response information fragments based on the order of the keywords.

9. The large-scale security management method based on organizational isolation and access control according to claim 1, characterized in that, The method further includes: When unauthorized content is detected in the response information, the field corresponding to the unauthorized content in the response information is hidden, and then the response information is sent to the business system.

10. The large-scale security management method based on organizational isolation and access control according to claim 1, characterized in that, Sending the response information to the business system includes: Add applicable boundaries to the response information and send the response information to the business system.

Images