Improved Text Emotion Prediction Using Combined Valence and Arousal Ordinal Classification

TL;DR

The paper tackles making emotion prediction from text more perceptually aware by treating emotions as ordinal rather than purely discrete categories. It first establishes a RoBERTa-CNN baseline and then introduces ordinal classification along a valence scale, followed by a two-dimensional ordinal approach in the valence-arousal space. The results show that ordinal methods preserve accuracy while reducing severe misclassifications, and the 2D approach significantly improves performance on 23-emotion GoEmotions data and enables localization near ground truth even for unseen emotions. This approach offers a more nuanced and robust framework for emotion representation in text, with potential benefits for downstream systems like TTS and dialogue agents.

Abstract

Emotion detection in textual data has received growing interest in recent years, as it is pivotal for developing empathetic human-computer interaction systems. This paper introduces a method for categorizing emotions from text, which acknowledges and differentiates between the diversified similarities and distinctions of various emotions. Initially, we establish a baseline by training a transformer-based model for standard emotion classification, achieving state-of-the-art performance. We argue that not all misclassifications are of the same importance, as there are perceptual similarities among emotional classes. We thus redefine the emotion labeling problem by shifting it from a traditional classification model to an ordinal classification one, where discrete emotions are arranged in a sequential order according to their valence levels. Finally, we propose a method that performs ordinal classification in the two-dimensional emotion space, considering both valence and arousal scales. The results show that our approach not only preserves high accuracy in emotion prediction but also significantly reduces the magnitude of errors in cases of misclassification.

Figures (4)

• Figure 1: ISEAR Emotions Valence Order
• Figure 2: Confusion Matrices on ISEAR dataset
• Figure 3: Error histograms of models trained for ordinal and baseline (softmax) classification on ISEAR dataset.
• Figure 4: The emotions grid, as described by Russel. In pink color depicted the distribution of joy emotion, which was not seen during training.