Enhancing Thai Rice Query Assistance through a Knowledge-Driven Approach Using GraphRAG

Abstract

Thai rice farmers face significant challenges accessing timely and accurate information for crucial decisions regarding variety selection, soil management, and adapting to climate change. While Retrieval-Augmented Generation (RAG) systems aim to provide information, traditional RAG often struggles with complex queries requiring interconnected knowledge and can yield generic or less relevant answers in specialized domains like agriculture due to its reliance on the semantic similarity of isolated text chunks. This paper introduces and evaluates GraphRAG, a knowledge graph-enhanced RAG approach, designed specifically to overcome these limitations and improve query assistance for Thai rice cultivation. The methodology involves constructing a knowledge graph from key Thai rice farming documents and integrating it with a large language model to provide context-aware responses, comparing its performance against a traditional RAG baseline. Results demonstrate GraphRAG’s superior effectiveness; user preference tests showed participants favored GraphRAG responses (52.9%) significantly more than traditional RAG (35.3%), particularly for complex queries requiring nuanced understanding. Quantitatively, GraphRAG showcased enhanced efficiency, reducing the average query response time by nearly 3 times (from 1.43 seconds for RAG to 0.41 seconds) and decreasing memory usage by over 50% (from 457.42 KB for RAG to 213.09 KB). This study concludes that GraphRAG offers a valuable approach for enhancing information retrieval accuracy, contextual understanding, and system efficiency in specialized, low-resource agricultural domains, highlighting its significance for providing better decision support to farmers.