This project explores the extraction of structured knowledge from Large Language Models (LLMs) using Retrieval-Augmented Understanding (RAU) and Knowledge Graphs (KGs). The aim is to transform unstructured textual information into structured knowledge representations that enable:
-
Efficient querying
-
Interpretability
-
Intelligent reasoning
A Graph Kernel Mean Embedding technique is applied to measure semantic similarity between knowledge pathways and to evaluate the importance of graph components such as edges and nodes. The project also integrates uncertainty quantification by assessing relationship confidence and structural significance.
Key Applications
-
Biomedical knowledge extraction
-
Automated fact-checking
-
Decision support systems
-
Knowledge-driven question answering
Objectives
-
Develop an automated LLM-to-KG pipeline for extracting and structuring knowledge
-
Perform embedding-based graph analysis to evaluate relational importance
-
Implement an uncertainty quantification module to assess confidence in extracted knowledge
-
Conduct benchmark evaluations and comparative studies with existing extraction techniques
