SentiANNO: Annotating Sentiment in Austrian Historical Newspapers

This contribution presents SentiANNO, a sentiment-annotated corpus from historical Austrian newspapers; and describes the annotation process and the training of annotators. The corpus, which covers journalistic texts in German from 1800 to 1938, addresses the lack of sentiment-annotated resources for historical newspapers. Such a resource can be used to fine-tune existing Machine Learning models for Sentiment Analysis.The corpus includes texts from ANNO and DIGITARIUM collections, categorized into four sentiment categories (positive, negative, neutral and mixed) by three non-expert annotators. Throughout the annotation process, Doccano proved to be the most effective annotation tool, with preliminary results showing over 70% inter-annotator agreement despite genre and language complexity. The corpus will be publicly available on Zenodo, supporting open access.