Knowledge Graph Creation Using Embeddings and LLM-Based Association
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The development of ontologies for knowledge graphs is resource-intensive, requires expertise, and struggles with ambiguous domain languages and the challenge of capturing tacit knowledge, leading to outdated and inefficient knowledge graph creation.
Innovation Solution
A system combining embedding models and large language models (LLMs) to identify classes and relationships without prior ontology development, leveraging their capabilities to create knowledge graphs by generating embeddings and determining associations between entities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If ontology development is used for knowledge graph creation, then the knowledge graph can capture domain knowledge and relationships, but the process becomes resource-intensive and requires expert knowledge
Solution Approach 1:
The patent replaces the manual ontology development process with automated machine learning models. Specifically, embedding models generate vector representations of entities, and large language models determine relationships between them, eliminating the need for manual expert curation while maintaining knowledge graph accuracy
Solution Approach 2:
The system enables self-service knowledge graph creation by automatically generating embeddings and determining relationships without requiring domain experts. The machine learning models autonomously process unstructured data and construct the knowledge graph structure, making the process accessible without specialized knowledge
2Loss of information
If traditional ontology-based methods are used, then domain expertise can be captured, but the process is time-consuming and outdated
Solution Approach 1:
The patent performs preliminary action by pre-computing embeddings for all entities in the dataset before relationship determination. This allows the system to efficiently query and compare entity representations when building relationships, significantly reducing the time required for knowledge graph construction while capturing tacit knowledge from unstructured data
Solution Approach 2:
Manual ontology development is replaced with automated ML pipelines that process unstructured data through embedding models and language models, dramatically reducing creation time while capturing domain knowledge that would otherwise require extensive manual expert input
3Measurement precision
If manual ontology development is performed, then precise domain relationships can be defined, but it requires significant expert resources and is not scalable
Solution Approach 1:
The patent substitutes manual ontology development with automated machine learning systems. Embedding models generate precise vector representations that capture semantic relationships, and large language models determine relationship types with high precision, achieving both accuracy and scalability simultaneously
Solution Approach 2:
The system changes the parameter of relationship definition from manual expert judgment to automated model-based determination. By adjusting model parameters such as embedding dimensions and similarity thresholds, the system maintains relationship precision while dramatically increasing creation speed and scalability
Data Source
AI summary
Certain aspects of the disclosure provide techniques for creating a knowledge graph. A method generally includes for each respective item, of a plurality of items, associated with a respective industry: adding an item node in the knowledge graph for the respective item; adding an industry node in the knowledge graph for the respective industry if no industry node for the respective industry exists in the knowledge graph; generating semantically similar items to the respective item; prompting one or more machine learning models to determine that the respective item and at least one semantically similar item of the set of semantically similar items are associated; and generating an edge between the respective item and the at least one semantically similar item in the knowledge graph based on the association determination.