A Comparative Study of Advanced Retrieval-Augmented Generation Methods for Educational Content Creation
Ravi Bhargav, Sushama Wagh, M Parimi, Shamita Dhamankar, Shashank S. V
Problems Identified (4)
RAG hallucination and factuality: Large language model outputs need grounding in external knowledge to reduce hallucinations and improve factual correctness.
Vector retrieval semantic limitations: Traditional vector-based retrieval methods struggle to capture semantic relationships, limiting generated response depth.
RAG hallucination and factuality: Large language model outputs need grounding in external knowledge to reduce hallucinations and improve factual correctness.
Vector retrieval semantic limitations: Traditional vector-based retrieval methods struggle to capture semantic relationships, limiting generated response depth.
Proposed Solutions (5)
Comparative advanced RAG study: The paper compares two advanced RAG variants for educational content creation.
Graph RAG: Graph RAG uses structured graph-based knowledge representations to capture relationships between entities.
Fusion Retrieval: Fusion Retrieval combines BM25 keyword search with vector-based similarity search.
Comparative advanced RAG study: The paper compares two advanced RAG variants for educational content creation.
Graph RAG: Graph RAG uses structured graph-based knowledge representations to capture relationships between entities.
Results (3)
Control Systems query evaluation:
Fusion Retrieval F1 improvement:
Control Systems query evaluation:
Research Domain
Retrieval-Augmented Generation for educational content creation