A Knowledge Graph-based RAG for Cross-Document Information Extraction

Extracting information from multiple complex documents, such as research papers is challenging due to LLMs’ limited context size and their difficulty linking distant concepts. To address this, A propose a Knowledge-Graph-based Retrieval-Augmented Generation (RAG) system that builds a knowledge graph from research papers and uses task-specific retrieval methods—combining vector and graph search. This approach not only enhances retrieval but also provides a structured, interpretable representation of information. Our results show that the structured search method outperforms traditional techniques in managing cross-document context, leading to more effective and detailed extraction, and ultimately aiding the comprehension of complex documents.