EmotivITA includes two tasks, both constraint and unconstraint. In proposing these tasks, we aim to promote dimensional emotion analysis, a problem who has received increasing attention within the field of sentiment analysis in the English-speaking community, but not yet so among the Italian speakers.
Task A: Dimensional emotion regression
Prediction of Valence, Arousal and Dominance values based on a set of Italian sentences and annotations, using only the target annotated dimension for training.
Task B. Multi-dimensional emotion regression
Prediction of Valence, Arousal and Dominance values based on a set of Italian sentences and annotations, using all mentioned dimensions for training (so to exploit possible correlations within them, see below).
See our evaluation page for more information.
Task B intend to study and exploit potential correlations between Valence, Arousal and Dominance, which have been discussed in literature (Kuppens et al., 2012; Warriner et al., 2013).
P. Kuppens, F. Tuerlinckx, J. Russell, L. Barrett, The Relation Between Valence and Arousal in Subjective Experience, in «Psychological Bulletin», n. 4, vol. CXXXIX, 2012, pp. 917-940.
A.B. Warriner, V. Kuperman, M. Brysbaert, Norms of valence, arousal, and dominance for 13,915 English lemmas, in «Behavior Research Methods», n. 4, vol. XLV, 2013, pp. 1191–1207.