Non-discrete Sentiment Dataset Annotation: Case Study for Lövheim Cube Emotional Model

Communications in Computer and Information Science. 2020. Vol. 1242. P. 154–164.

Kolmogorova A.

The research project we are conducting is devoted to text emotional analysis. In this paper, we report the preliminary results of the non-discrete data assessment method, which uses an original interface developed to annotate texts according to emotion model known as Lövheim Cube. Swedish neurophysiologist H. Lövheim put eight basic emotions in the cube vertices according to the particular combination of three monoamines triggers each of them. We took four supporting diagonals of the cube and mapped them onto assessment scales: Distress/Enjoyment, Rage/Disgust, Shame/Excitement, Fear/Surprise. 172 human assessors were asked to adjust the pointer of a slider between two opposite emotions on the scales after having been read each of 48 text fragments retrieved from Russian social network VKontakte. By converting labeled scalars into spatial coordinates in the cube space, we obtained a set of comparable evaluations. The effectiveness of the approach has been validated using the Intra-class correlation metric. The proposed method offers noticeable benefits when compared to the discrete assessment procedure, giving to each text a multidimensional evaluation, which is closer to the natural text perception while reading.

Keywords: эмоции, emotion, Text annotation, аннотирование текста, сентимент анализ, Sentiment Analyses