Lövheim Cube-Backed Emotion Analysis: From Classification to Regression

Communications in Computer and Information Science. 2022. No. 1503. P. 97–107.

Kolmogorova A.

Nowadays sentiment and emotion analyses are widespread methodologies. However, most of all related tasks in classification manner use discrete classes as target variables: Positive vs Negative (sometimes accompanied by Neutral class), or discrete emotion classes (as Anger, Joy, Fear, etc.). Nonetheless, it is more likely that emotion is not discrete. In this paper, we argue that regression is more natural way to evaluate and predict emotions in text and apply regression framework in study of using Lövheim Cube emotional model for emotion analysis. A regression approach for predicting a point in 3-d space or a configuration of its diagonals can provide us with detailed analytics from an emotional diversity perspective. The preliminary results on regression values prediction performed by five different models demonstrate the need of optimization in regard to a precision. The additional conclusion is that the accuracy of classification is not affected significantly by the target variable type.

Keywords: regression model, Sentiment Analyses, Classification model