Structure and Diversity Aware Context Bubble Construction for Enterprise Retrieval Augmented Systems

This paper introduces a novel approach for constructing "context bubbles" in enterprise retrieval-augmented generation (RAG) systems, focusing on both the structural integrity and semantic diversity of retrieved information. The method aims to enhance the relevance and comprehensiveness of generated responses for enterprise applications by intelligently curating context, thereby improving decision-making and information access within large organizations.