EmotivITA: Dimensional and Multi-dimensional Emotion Analysis in Italian Texts: Guidelines

This document provides a detailed presentation of the EmotivITA task, its state of the art, the methodology for evaluating participating systems, and guidelines for participation. Various aspects of sentiment analysis have already been addressed in previous editions of Evalita, but EmotivITA introduces, for the first time, the problem of emotion regression according to the Valence, Arousal, Dominance model proposed by Russell and Mehrabian for emotion evaluation. The task consists of a monodimensional and a multidimensional subtask, and for both it will be possible to present constrained and unconstrained runs. Participating systems will be evaluated based on the most commonly used correlation metrics in the field, Pearson's r and Mean Absolute Error. EmotivITA aims to promote research on dimensional emotion analysis in the Italian language and contributes to the lively debate within the Natural Language Processing community regarding affective computing.