SENTIMAP is a spatiotemporal mapping system that visualizes the emotional contours of Istanbul from 1970 to 2023 by analyzing a large corpus of historical newspaper articles using zero-shot emotion classification with large language models (LLMs). The project geolocates emotional expressions across districts and periods, constructing a dynamic affective cartography of the city.
While sentiment analysis has proven effective in domains where emotional valence is explicit—such as social media or consumer reviews—its application to archival news presents unique challenges. Emotions in newspapers are rarely direct; they are embedded in narrative framing, lexical nuance, and emphasis. These subtleties are especially difficult to parse in morphologically rich languages like Turkish, where agglutination, flexible word order, and a lack of labeled emotional datasets hinder traditional NLP approaches.
SENTIMAP addresses these complexities through LLMs capable of capturing context, handling morphological variation, and identifying multiple co-occurring emotions without the need for extensive training data. This enables a granular, temporally sustained analysis of how collective emotions respond to crises, reforms, elections, and ideological shifts over time.
In Türkiye’s context—marked by media censorship and narrative control—public emotion becomes not only an expression but also a target and a byproduct of political control. The project responds to these conditions by repurposing the very tools often associated with mass influence and transforming them into instruments of counter-analysis.
The project is presented as a two-channel video installation: one screen animates the month-by-month emotional evolution of Istanbul over 54 years, while the other displays spatial and periodic infographics of emotional intensities. Blending computational analysis with conceptual inquiry, SENTIMAP reveals emotion as both a historical trace and a political signal, inviting viewers to navigate the city’s affective history through non-linear, exploratory encounters.