Automating Historical Insight Extraction from Large-Scale Newspaper Archives via Neural Topic Modeling

Automating Historical Insight Extraction from Large-Scale Newspaper Archives via Neural Topic Modeling Keerthana Murugaraj ,1,∗Salima Lamsiyah ,1 Marten During 2 and Martin Theobald 1 1Department of Computer Science, University of Luxembourg, 2, place de l’Universit´e, 4365, Esch-Belval Esch-sur-Alzette, Luxembourg and 2Centre for Contemporary & Digital History (C2DH), University of Luxembourg, 11 Porte des Sciences, 4366, Esch-Belval Esch-sur-Alzette, Luxembourg ∗Corresponding author. keerthana.murugaraj@uni.lu Abstract Extracting coherent and human-understandable themes from large collections of unstructured historical newspaper archives presents significant challenges due to topic evolution, Optical Character Recognition (OCR) noise, and the sheer volume of text. Traditional topic-modeling methods, such as Latent Dirichlet Allocation (LDA), often fall short in capturing the complexity and dynamic nature of discourse in historical texts. To address these limitations, we employ BERTopic. This neural topic-modeling approach leverages transformer- based embeddings to extract and classify topics, which, despite its growing popularity, still remains underused in historical research. Our study focuses on articles published between 1955 and 2018, specifically examining discourse on nuclear power and nuclear safety. We analyze various topic distributions across the corpus and trace their temporal evolution to uncover long-term trends and shifts in public discourse. This enables us to more accurately explore patterns in public discourse, including the co-occurrence of themes related to nuclear power and nuclear weapons and their shifts in topic importance over time. Our study demonstrates the scalability and contextual sensitivity of BERTopic as an alternative to traditional approaches, offering richer insights into historical discourses extracted from newspaper archives. These findings contribute to historical, nuclear, and social-science research while reflecting on current limitations and proposing potential directions for future work. Key words: Historical Text Mining, Unsupervised Automated Analysis, Neural Topic Modeling, BERTopic, Nuclear Discourse, Temporal Text Evolution 1. Introduction Newspapers have long been a vital primary source for historians, offering key insights into past events, societal trends, and cultural narratives. They help establish historical timelines, verify facts, and identify trends, providing detailed accounts of people, places, and events. Alongside other primary sources, newspapers continue to shape our understanding of history through rich first- hand documentation (Tibbo, 2002; Sun and Qin, 2025). Traditional methods of analyzing newspapers, such as manual reading and bibliographic approaches relying on subjective expertise, have notable drawbacks, especially when dealing with large volumes of unstructured textual archives. These methods can be time-consuming, prone to human bias, and inefficient for organizing and analyzing extensive collections of documents. To address these challenges, computational approaches have been increasingly employed in recent studies to interpret factual content and perform text classification tasks (Huang et al., 2022; Tsai et al., 2025). In contrast to traditional methods, text mining techniques based on Natural Language Processing (NLP)—particularly topic modeling—offer a scalable and automated solution for uncovering latent patterns within large corpora, without the need for manual categorization (Fridlund and Brauer, 2013; Vorobeva et al., 2025; Cvrˇcek and Berrocal, 1 arXiv:2512.11635v1 [cs.CL] 12 Dec 2025 2 Keerthana Murugaraj et al. 2025). Topic modeling identifies latent topics within vast document collections, thereby providing a more efficient, objective, and scalable solution to the challenges historians face in organizing and analyzing their research (Yang et al., 2011; Yamashita and Uchida, 2025) In this paper, we explore the effectiveness of BERTopic (Grootendorst, 2022), a state-of-the-art neural-based topic-modeling technique, in analyzing historical newspaper archives. It leverages transformer- based embeddings (Vaswani et al., 2017) to capture nuanced relationships between words and topics, outperforming classical topic-modeling methods such as Latent Dirichlet Allocation (LDA) (Blei et al., 2003), probabilistic Latent Semantic Analysis (pLSA), and Non-negative Matrix Factorization (NMF) (Lee and Seung, 1999). By using pre-trained language models, BERTopic has demonstrated superior performance in terms of generating coherent and human-understandable topics, even in large and complex document collections (Medvecki et al., 2023; Mutsaddi et al., 2025). The contextualized embedding-based representations enable the model to better understand subtle thematic shifts over time, a feature particularly useful for analyzing historical data where linguistic and contextual variations are significant. Our work builds upon the foundations lai

This content is AI-processed based on open access ArXiv data.