Researchers use advanced LLMs and topic modeling to analyze 20th-century Slovene newspapers and uncover ideological differences in historical public discourse.

Study analyzes Slovenec and Slovenski narod newspapers from the sPeriodika corpus using BERTopic for topic modeling and LLM-based sentiment analysis

Identifies clear ideological differences between the two publications, reflecting their conservative-Catholic and liberal-progressive orientations respectively

Evaluates four instruction-following LLMs for sentiment classification in OCR-degraded historical Slovene text and selects GaMS3-12B-Instruct as the most effective model

Combines computational methods with qualitative discourse analysis to examine representations of collective identity, political orientation, and national belonging at the turn of the 20th century