Automatically Detecting Political Viewpoints in Norwegian Text

Abstract

We introduce three resources to support research on political texts in Scandinavia. The encoder-decoder transformer models sp-t5 and sp-t5-keyword were trained on political texts. The nor-pvi (available at https://tinyurl.com/nor-pvi) data set comprises political viewpoints, stances, and summaries for Norwegian. Experiments with four distinct tasks show that large-scale models, such as nort5 perform slightly better. Still, sp-t5 and sp-t5-keyword perform almost on par and require much less data and computation.