Description
The ongoing digitization of historical newspapers offers the opportunity to study changes in ‘the nation’s mood’ in a large scale. Although sentiments form a vital aspect of this, extracting them from the newspaper text is not a trivial task. Explicit manifestations of sentiments in newspaper data are highly dependent, among other things, on the function of the text (news vs. opinion), the self-identification of the newspaper (high-brow vs. popular) and the historical context. This project aims to address this challenge by applying and evaluating multiple approaches for sentiment analysis. It will compare machine learning with dictionary based approaches and historical with present-day sentiments – with the aim of setting new standards for the analysis of sentiments in factual texts where sentiments are expressed implicitly.