SysML Knowledge Graph Query Mapping for Accurate Natural Language Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems engineering models, such as SysML, lack efficient methods for bridging natural language queries with structured knowledge graphs, leading to high computational costs and unreliable results due to the need for real-time schema discovery or loss of semantic information.
Innovation Solution
A hybrid approach combining vector semantic search with pre-defined template queries to translate SysML models into Knowledge Graphs, leveraging domain-specific structural knowledge, and integrating Large Language Models (LLMs) for question-answering over the graph model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LLMs dynamically discover and understand graph schemas before generating queries, then query accuracy is improved, but computational cost increases significantly
Solution Approach 1:
The patent pre-generates query templates based on the graph schema before runtime, storing them for later use. This preliminary action eliminates the need for LLMs to dynamically discover schemas during query generation, reducing computational cost while maintaining query accuracy through pre-validated template structures.
Solution Approach 2:
The patent creates template queries that replicate common query patterns based on the graph schema. Instead of generating unique queries each time, the system copies and adapts pre-defined templates, significantly reducing the computational burden on LLMs while preserving query effectiveness.
2Speed
If pure vector search is used without leveraging structural relationships, then query processing speed is improved, but semantic information is lost
Solution Approach 1:
The patent segments the query processing into two parts: vector search for initial candidate retrieval (maintaining speed) and graph template matching for precise semantic querying (preserving information). This segmentation allows each method to operate in its optimal domain without sacrificing overall performance or information integrity.
Solution Approach 2:
The patent introduces query templates as an intermediary between vector search and graph querying. The templates serve as a bridge that incorporates structural relationships into the search process, allowing vector search to operate efficiently while ensuring semantic information is preserved through template-guided refinement.
3Adaptability or versatility
If standardized languages are adopted across industries, then interoperability is improved, but the universe of individuals who understand these languages remains limited
Solution Approach 1:
The patent introduces natural language processing and LLMs as intermediaries between users and standardized graph query languages. Users can interact with systems using natural language while the system translates these into standardized graph queries, maintaining interoperability through standard protocols while dramatically improving accessibility for non-experts.
Solution Approach 2:
The patent uses pre-defined query templates that capture common inquiry patterns. These templates are copied and adapted to handle diverse user needs while maintaining consistency with standardized graph schemas, enabling broad interoperability through reusable patterns that don't require deep domain knowledge.
Data Source
AI summary
A process and system for facilitating natural language interrogation of system engineering models is described. The process and system utilize multiple formatting translations of the original SysML model into a graph schema which can be queried by a user using LLM-backed Retrieval Augmented Generation (RAG).