Fine-grained extraction of geospatial and temporal information from Chinese historical newspapers

Abstract Historical newspapers are invaluable repositories of comprehensive knowledge, capturing the essence of diverse societal shifts and pivotal events across varying epochs. By scrutinizing and identifying intricate details such as place names, locations, dates, and a diverse array of Points of Interest spanning global, regional, and local scales, including countries, cities, buildings, streets, monuments, and forests, historical newspapers facilitate the reconstruction of spatial distributions and timelines of past events. This study proposes a sophisticated multi-tiered geospatial and temporal information framework. This framework is exemplified through empirical research utilizing historical newspaper texts from Chinese ‘Shengjing Times Changchun Compilation’. Leveraging advanced deep learning models such as BiLSTM, BERT, and Boundary Smoothing for meticulous data annotation and extraction, the study demonstrates the feasibility and effectiveness of extracting geospatial and temporal information from historical newspaper texts. The outcomes of this research offer invaluable methodological insights and guidance for contributing significantly to the field of historical studies and information retrieval.