Universal Knowledge Graph Querying With LLM-Guided Data Discovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The abundance of data in data sources and the sheer size of data models creates difficulty for users when searching for relevant information, making it impractical for them to manually determine where to begin searching in large databases to maximize their search effectiveness.
Innovation Solution
A system that generates a universal knowledge graph based on multiple local knowledge graphs, fine-tunes a large language model (LLM) on this graph, and processes prompts to generate responses to queries, utilizing retrieval augmented generation (RAG) techniques to provide insights and recommendations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If users manually search in large databases to find relevant information, then they can access detailed data, but the time and effort required increases significantly
Solution Approach 1:
The system pre-generates knowledge graphs from available data sources before users need to query them. These knowledge graphs organize data relationships in advance, allowing the LLM to quickly retrieve and process relevant information without users having to manually search through raw databases, thus reducing search time while maintaining information accessibility
Solution Approach 2:
The patent introduces an LLM-based intermediary system that sits between users and the underlying data sources. The LLM processes natural language queries, translates them into appropriate search operations on the knowledge graphs, and presents results in an accessible format. This intermediary handles the complexity of data retrieval, allowing users to access information quickly without needing to understand the underlying database structures
2Adaptability or versatility
If the system provides comprehensive data access, then users can find relevant information, but the complexity of the system increases
Solution Approach 1:
The system creates simplified copies of the underlying data structures in the form of knowledge graphs. Instead of requiring users to interact with complex raw databases, the LLM works with these simplified graph representations that capture essential relationships. This copying approach maintains data access capability while reducing the apparent complexity users must deal with
Solution Approach 2:
The LLM is trained to handle multiple types of queries and data sources through a single unified interface. The knowledge graph structure provides a universal representation that can accommodate different data types and relationships. This universality allows the system to provide comprehensive data access across multiple sources without increasing the complexity of the user interface or query mechanism
3Productivity
If the system processes more data sources, then the value of business intelligence increases, but the difficulty of managing and integrating data increases
Solution Approach 1:
The system segments the integration process by creating separate knowledge graphs for different data sources while maintaining a unified LLM interface. Each knowledge graph can be independently constructed and updated from its corresponding data source, allowing the system to process multiple data sources without requiring complex end-to-end integration logic. The LLM orchestrates queries across these segmented graphs, enabling high business intelligence value while managing integration complexity through modular design
Data Source
AI summary
A system and method for providing query responses from a big data system utilizing a knowledge graph is presented. The method includes generating a plurality of local knowledge graphs, each local knowledge graph of the plurality of local knowledge graphs generated respective of a unique plurality of data sources; generating a universal knowledge graph based on the generated plurality of local knowledge graphs; fine-tuning a large language model (LLM) based on the generated universal knowledge graph; receiving a query directed at a data source of the plurality of unique data sources of a first local knowledge graph; generating a prompt for the LLM based on the received query; and processing the prompt utilizing the LLM to generate an output, wherein the output is a response to the received query.