A Preliminary Study of RAG for Taiwanese Historical Archives

The recent preliminary study on Retrieval-Augmented Generation (RAG) applied to Taiwanese Historical Archives marks a significant step in utilizing AI for historical research. By examining datasets from Fort Zeelandia and the Taiwan Provincial Council Gazette, the study reveals that integrating metadata at early stages can enhance both retrieval and answer accuracy. However, it also uncovers persistent challenges faced by RAG systems, such as hallucinations during generation and difficulties in addressing complex temporal or multi-hop historical queries. This research not only contributes to the understanding of RAG's effectiveness in knowledge-intensive tasks but also emphasizes the need for further exploration and refinement of these systems to improve their reliability and performance in historical contexts.