Enhancing Vector based Retrieval Augmented Generation with Contextual Knowledge Graph Construction

The proliferation of unstructured text data necessitates efficient information retrieval systems. Traditional vector-based Retrieval Augmented Generation (RAG) models often fail to capture complex relationships and contextual nuances, limiting effectiveness in knowledge-intensive tasks. We introduce Contextual Knowledge Graph Construction (CKGC), a novel approach enhancing vector-based RAG by dynamically building a knowledge graph that reflects inherent data structures and connections.CKGC leverages text chunking, large language models (LLMs), and ontology mapping. By segmenting text and using LLMs to identify key entities and relationships, CKGC constructs a contextualized knowledge graph enriching information representation. This bridges the gap between semantic similarity and deeper contextual understanding, enabling more accurate and nuanced retrieval.Experiments on 2,000 lease agreements demonstrate that CKGC significantly improves vector-based RAG in information retrieval and question answering tasks, with substantial gains in Mean Reciprocal Rank (MRR) and Top-k Accuracy. CKGC’s adaptability across domains positions it as a valuable tool for enhancing performance and understanding of complex textual data. Our findings underscore CKGC’s transformative potential in unlocking insights from vast text corpora, paving the way for more intelligent and context-aware information retrieval systems.