SICS: Valence annotation based on seeds in word space

This paper reports on a experiment to identify the emotional loading (the “valence”) of news headlines. The experiment reported is based on a resource-thrifty approach for valence annotation based on a word-space model and a set of seed words. The model was trained on newsprint, and valence was computed using proximity to one of two manually defined points in a high-dimensional word space — one representing positive valence, the other representing negative valence. By projecting each headline into this space, choosing as valence the similarity score to the point that was closer to the headline, the experiment provided results with high recall of negative or positive headlines. These results show that working without a high-coverage lexicon is a viable approach to content analysis of textual data.