Knowledge Graph Generation With SQL-Friendly Semantic Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to allow users to query knowledge graphs using standard SQL while abstracting the complexities of graph traversal and semantic querying, leading to inefficiencies and a gap between SQL-based tools and the deeper relational insights captured in knowledge graphs.
Innovation Solution
A multi-agent pipeline utilizing ontology and mapping generation agents, supported by a data access agent, iteratively refines semantic representations and mapping files to convert natural language queries into accurate SQL queries, integrating domain-specific logic and leveraging Large Language Models (LLMs) for seamless interaction with data sources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If knowledge graphs are used to represent relationships between entities, then contextual richness and machine-readability are improved, but querying complexity increases due to requirement of SPARQL or graph-based query languages
Solution Approach 1:
The patent introduces an intermediary translation layer that converts SQL queries into graph-based queries (SPARQL or property graph queries). This mediator component allows users to interact with knowledge graphs using familiar SQL syntax while the system handles the complexity of graph traversal and semantic querying in the background, thus resolving the contradiction between contextual richness and querying ease.
2Ease of operation
If SQL-like interfaces are provided over knowledge graph engines, then user familiarity is improved, but full semantics of the underlying graph cannot be captured
Solution Approach 1:
The translation layer acts as an intelligent intermediary that not only translates SQL syntax but also enriches queries by inferring semantic relationships from the knowledge graph schema. It adds graph-specific capabilities such as path traversal, entity resolution, and relationship inference while maintaining SQL syntax, thus capturing full graph semantics without sacrificing user familiarity.
Solution Approach 2:
The system replaces the mechanical, syntax-only SQL processing with a semantic-aware translation mechanism that understands both SQL syntax and graph semantics. This substitution allows the system to interpret SQL queries in the context of the knowledge graph structure, enabling rich semantic capture while maintaining the simplicity of SQL interface.
3Adaptability or versatility
If data is duplicated between relational and graph systems, then query flexibility is improved, but maintenance overhead and inconsistencies increase
Solution Approach 1:
The patent extracts the graph data model from a separate duplication system and integrates it as a semantic layer over the existing relational database. Instead of maintaining duplicate data in both relational and graph systems, the solution uses the relational database as the single source of truth and builds the knowledge graph semantics on top of it through schema inference and ontology generation, thus eliminating maintenance overhead while preserving query flexibility.
Data Source
AI summary
A method (400) and system (100) of generating knowledge graph of the data repository is disclosed. The method (400) includes receiving input data (302) and access of data repository (304). The method (400) may include generating semantic (310) representation of data repository (304) schema based on input data (302) and data repository (304) using language model. The method (400) may further include validating semantic representation (310) syntactically and with respect to input data (302). The method (400) may further include generating mapping (320) file of data repository (304) schema based on semantic representation (310) and data repository (304) using language model. The mapping file (320) may include mapping of plurality of elements of semantic representation (310) to corresponding elements in input data (302). Further, the method (400) includes validating mapping file (320) syntactically and semantically based on semantic representation (310), data repository (304) and input data (302).