LLM Query Translation for Ambiguity-Resilient Knowledge Graph Access
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Querying knowledge graph databases is cumbersome for users due to the need to learn database query languages specific to each type of database, which is exacerbated by large ontologies and the risk of null responses from misunderstandings in query formulation.
Innovation Solution
A system leveraging large language models (LLMs) to process natural language queries, correcting ambiguities, generating compatible database queries, and handling errors to provide seamless access to knowledge graph databases without requiring coding skills.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If users query knowledge graph databases directly, then query access is straightforward, but users must learn database query languages specific to each database type which increases operational complexity
Solution Approach 1:
The system introduces an intermediary layer consisting of a large language model and query translator between the user's natural language query and the knowledge graph database. This intermediary converts user-friendly natural language queries into database-specific query languages, eliminating the need for users to learn complex query syntax while maintaining direct access to the database.
Solution Approach 2:
The system changes the parameter of query language from technical database-specific syntax to natural human language. By transforming the query interface parameter, the system makes query formulation accessible to users without coding knowledge while preserving the ability to access structured data in knowledge graph databases.
2Quantity of substance
If the knowledge graph database ontology is made comprehensive, then data coverage is improved, but query formulation becomes more error-prone and complex
Solution Approach 1:
The system implements feedback mechanisms where the large language model analyzes query results and user interactions to refine query formulations. When queries return null or unexpected results, the system can iteratively adjust the query based on feedback, improving accuracy without requiring users to manually refine complex queries.
Solution Approach 2:
The query translation system performs self-service by automatically adapting to different ontology structures and query contexts. The large language model autonomously handles the complexity of comprehensive ontologies, resolving ambiguities and formulating accurate queries without user intervention in the translation process.
3Adaptability or versatility
If multiple knowledge graph database types are supported, then system versatility is improved, but query language complexity increases
Solution Approach 1:
The system creates a universal query interface that works across multiple knowledge graph database types through a single natural language input mechanism. The large language model and query translator serve multiple functions by adapting to different database schemas and query languages, allowing one interface to handle diverse database types without increasing user burden.
Data Source
AI summary
Aspects of the subject disclosure may include, for example, obtaining a user query to access information from a group of knowledge graph databases, the user query corresponding to user-generated input, and identifying a group of terms within the user query that reduces a likelihood of identifying a match within the group of knowledge graph databases. Further embodiments can include adjusting the user query resulting in a first adjusted user query that enables identifying a knowledge graph database from the group of knowledge graph databases, identifying the knowledge graph database from the group of knowledge graph databases based on the first adjusted user query, and generating a first database query based on the first adjusted user query and the knowledge graph database, in which the first database query would have a computer-readable syntax compatible with the identified knowledge graph database. Other embodiments are disclosed.