APPLYING LLAMAINDEX AND GRAPHRAG FOR KNOWLEDGE GRAPH CONSTRUCTION AND QUERYING

Abstract

This paper presents a method that combines LLaMA Index and GraphRAG to construct and query knowledge graphs from plain-text data. We introduce an automated pipeline for entity and relation extraction as well as community-based summarization, integrating a large language model (LLM) with a community-detection algorithm. The resulting system supports efficient processing of natural-language queries. Experiments on a tourism dataset from Ho Chi Minh City demonstrate its ability to answer complex queries accurately. Empirical results show that our approach improves query-response effectiveness by approximately 4% compared with using an LLM alone. We also discuss challenges related to computational cost and the need for high-quality data, along with the method’s potential applications in knowledge management and semantic search.