This thesis focuses on designing an Agentic Graph Retrieval-Augmented Generation (RAG) system specifically for question answering in oral maxillofacial surgery (OMS) guidelines. By leveraging graph-based knowledge representation and advanced language models, the system aims to improve accuracy and efficiency in accessing and interpreting surgical guidelines. Key research areas include the integration of graph databases, ontology-based retrieval, and natural language processing to facilitate precise, contextually relevant responses to complex clinical queries.
Thesis Type |
|
Student |
Amelie Dittmann, Saif Nasir |
Status |
Running |
Presentation room |
Seminar room I5 6202 |
Supervisor(s) |
Stefan Decker |
Advisor(s) |
Yongli Mou |
Contact |
mou@dbis.rwth-aachen.de |
Background
Oral maxillofacial surgery involves specialized and intricate procedures that require precise guideline interpretation. With a vast amount of literature and clinical recommendations, it can be challenging for healthcare professionals to access concise, relevant information on demand. RAG systems enhance the retrieval process by combining the strengths of large language models and structured data sources, such as knowledge graphs. In this project, an agentic RAG approach will be applied, allowing the system to reason through guideline data by associating entities and relationships, thereby enabling more accurate information extraction and retrieval for specific OMS scenarios.
Objectives
- Develop an agentic graph RAG system tailored for OMS guideline question answering.
- Investigate the use of graph-based knowledge representation to improve retrieval accuracy.
- Enable ontology-based retrieval methods to enhance contextual understanding and precision.
- Optimize language model integration to support agentic reasoning capabilities for complex clinical questions.
Tasks
- Integrate graph databases with large language models to support OMS guideline retrieval.
- Build and populate knowledge graphs from OMS guidelines to represent surgical procedures, terminology, and decision pathways.
- Develop ontology modules to categorize and retrieve information in alignment with clinical queries.
- Evaluate system performance on typical OMS question-answering tasks, with metrics for accuracy, precision, and response time.
Deep Knowledge of Deep Learning, Large Language Models
Programming language – Python (PyTorch, Transformers, etc.)