Emotion analysis refers to the classification of text according to emotions, based on psychological models. It can be conducted in a supervised way using annotated datasets, or in an unsupervised way using available emotion lexicons. Annotating texts with emotions is a challenging task because emotions are often subtle and context-dependent, and the language used to express them varies across domains and cultures. This challenge is even greater for low-resourced languages such as Croatian. The seminar will present a methodology for the automatic extraction of emotion lexicons based on the measure of s-discordance from symbolic data analysis, and its application to semi-supervised classification of textual documents according to emotions in the vector space model. In addition, possible applications of symbolic data analysis will be discussed for the intrinsic evaluation of word representations in neural language models and for the classification of textual documents by emotions using language models.

Se realizará de manera virtual mediante la platadorma de zoom con la siguiente información:

ID de la reunión: 666 155 1739

PhD. Jasminka Dobša