Abstract
In this paper, we present the results of experiments aiming to validate a two-dimensional typology of affective states as a suitable basis for affective classification of texts. Using a corpus of English weblog posts, annotated for mood by their authors, we trained support vector machine binary classifiers to distinguish texts on the basis of their affiliation with one region of the space. We then report on experiments which go a step further, using four-class classifiers based on automated scoring of texts for each dimension of the typology. Our results indicate that it is possible to extend the standard binary sentiment analysis (positive/negative) approach to a two dimensional model (positive/negative; active/passive), and provide some evidence to support a more fine-grained classification along these two axes.
| Original language | English |
|---|---|
| Title of host publication | Sentiment and Subjectivity in Text, Workshop at the Annual Meeting of the Association of Computational Linguistics (ACL 2006) |
| Publication status | Published - 22 Jul 2006 |
| Event | Sentiment and Subjectivity in Text, Workshop at the Annual Meeting of the Association of Computational Linguistics (ACL 2006) - Sydney, Australia Duration: 22 Jul 2006 → … |
Workshop
| Workshop | Sentiment and Subjectivity in Text, Workshop at the Annual Meeting of the Association of Computational Linguistics (ACL 2006) |
|---|---|
| Period | 22/07/06 → … |