LLM Chat Server Catalog Segmentation for Data Security
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in securely utilizing large language models for enterprise data assets, leading to exposure of internal information and decreased response accuracy due to hallucination or false responses.
Innovation Solution
A chat service provision method using a large language model that applies a natural language conversation scheme to data assets, allowing users to request necessary data via natural conversation, while implementing permission management based on individual user workspaces and using a plug-in scheme to extend LLM services.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a large language model is applied to enterprise data assets to enable natural language queries, then user accessibility and ease of operation are improved, but internal information exposure and response accuracy deteriorate
Solution Approach 1:
The patent segments enterprise data assets into multiple catalogs organized by predetermined categories (e.g., financial data, HR data, operational data). Instead of providing the LLM access to all data assets at once, the system divides the data into manageable catalog units that can be selectively accessed based on user permissions and query context, thereby maintaining response accuracy while enabling natural language queries.
Solution Approach 2:
The patent introduces a catalog management system as an intermediary between the LLM and enterprise data assets. This intermediary layer filters and manages data access, ensuring that the LLM only processes relevant and authorized data portions. The catalog system acts as a mediator that preserves response accuracy by preventing information overload and unauthorized access while maintaining user accessibility through natural language interfaces.
2Loss of information
If the large language model accesses extensive enterprise data assets, then the comprehensiveness of information retrieval is improved, but information security and exposure risks worsen
Solution Approach 1:
The patent segments enterprise data assets into multiple catalogs organized by predetermined categories (e.g., financial data, HR data, operational data). Instead of providing the LLM access to all data assets at once, the system divides the data into manageable catalog units that can be selectively accessed based on user permissions and query context, thereby maintaining response accuracy while enabling natural language queries.
Solution Approach 2:
The patent applies different access control policies and security measures to different catalog units based on their sensitivity and importance. Critical data catalogs have stricter access controls, while less sensitive catalogs have more permissive access. This local differentiation of security measures ensures comprehensive information retrieval where appropriate while protecting sensitive information from exposure.
3Object-affected harmful factors
If permission management is implemented for each user workspace and plug-in, then information security is improved, but system complexity increases
Solution Approach 1:
The patent implements a universal permission management system that handles access control for multiple workspaces, catalogs, and plug-ins through a single integrated framework. Instead of implementing separate permission systems for each component, the catalog-based permission system provides multi-functional access control that works across all data assets and LLM interactions, reducing overall system complexity while maintaining security.
Solution Approach 2:
The patent establishes permission rules and catalog associations in advance, before users execute queries. The system pre-configures which users can access which catalogs and what operations they can perform. This preliminary setup of permissions and associations simplifies real-time query processing, as the system only needs to check pre-established rules rather than making complex authorization decisions for each individual query.
Data Source
AI summary
Disclosed are a chat service providing method that uses a large language model (LLM) and a chat service server using the same. A chat service provision method of a chat service server that uses a large language model according to an embodiment may include an operation of receiving a natural language-based query message from a user, an operation of searching for a target catalog corresponding to the query message from among a plurality of catalogs generated based on data assets, an operation of associating the target catalog with an online workspace assigned to the user, an operation of calling an LLM application that provides an LLM service corresponding to the query message in the workspace, and generating an inference result for the query message based on the target catalog by using the LLM application, and an operation of providing a response message including the inference result to the user.