A Study Case of Automatic Archival Research and Compilation using Large Language Models

Archival research and compilation is a specialized task that focuses on exploration, selection and processing of vast quantities of archival documents pertaining to specific subjects. Traditionally, this task has been characterized by its labor-intensive and time-consuming requirements. In recent years, the advancement of artificial intelligence has made automatic archival research and compilation tasks feasible. However, the limited availability of relevant samples imposes significant constraints on the application of deep learning models, given their high demand for sufficient data and knowledge. In this paper, we present a study case and propose an innovative method for automatic archival research and compilation, leveraging the robust knowledge base and text generation ability offered by large language models. Specifically, our method comprises three essential components: document retrieval, document summarization, and rule-based compilation. In the document summarization component, we leverage fine-tuned large language models to enhance the performance by simulation data generation and summary generation. Experimental results substantiate the effectiveness of our method. Furthermore, our method provides a general idea in using large language models, as well as a solution for addressing similar challenges in different domains.