Unlocking the Digitized Historical Newspaper Archive

This paper aims to utilize historical newspapers through the application of computer vision and machine/deep learning to extract the headlines and illustrations from newspapers for storytelling. This endeavor seeks to unlock the historical knowledge embedded within newspaper contents while simultaneously utilizing cutting-edge methodological paradigms for research in the digital humanities (DH) realm. We targeted to provide another facet apart from the traditional search or browse interfaces and incorporated those DH tools with place- and time-based visualizations. Experimental results showed our proposed methodologies in OCR (optical character recognition) with scraping and deep learning object detection models can be used to extract the necessary textual and image content for more sophisticated analysis. Timeline and geodata visualization products were developed to facilitate a comprehensive exploration of our historical newspaper data. The timeline-based tool spanned the period from July 1942 to July 1945, enabling users to explore the evolving narratives through the lens of daily headlines. The interactive geographical tool can enable users to identify geographic hotspots and patterns. Combining both products can enrich users’ understanding of the events and narratives unfolding across time and space.