SENTIMAP: spatiotemporal mapping of emotions in historical newspapers using LLMs

Abstract

SENTIMAP is a data-driven visualization project that maps historical emotional trends across Turkish urban districts from 1970 to 2024, generating district-level maps where colors reflect emotional shifts over time. It simultaneously visualizes periodic emotional summits, emotional apexes, and aggregate emotional volume, offering a rich spatiotemporal understanding of how public feeling accumulates, intensifies, and transforms alongside socio-political developments. This enables a more nuanced reading of the emotional dimensions that unfold in tandem with historical change. It adopts a more robust and flexible approach by leveraging large language models (LLMs), which excel at capturing context, tone, and nuance within complex linguistic structures. Unlike conventional techniques that rely on custom rule sets, labeled training data, or rigid pipelines, LLMs generalize emotional understanding across varied historical and stylistic texts with minimal preprocessing. This results in a more scalable and accurate emotional extraction process - especially valuable when working with decades of unstructured, archival media. This approach is particularly significant for Turkish, a language that poses unique challenges for natural language processing (NLP) due to its agglutinative structure, extensive morphology, and complex grammar. These linguistic features often undermine the effectiveness of traditional emotional analysis methods such as lexicon-based scoring or statistical classifiers - challenges that LLMs are uniquely positioned to overcome. © 2025 Copyright held by the owner/author(s).