Are Emotions Arranged in a Circle? Geometric Analysis of Emotion Representations via Hyperspherical Contrastive Learning

TL;DR

This paper introduces a contrastive learning method to embed emotions in a circular, hyperspherical space within language models, enhancing interpretability but with trade-offs in high-dimensional and fine-grained tasks.

Contribution

It is the first to incorporate psychological circumplex models directly into deep language model embeddings using hyperspherical contrastive learning.

Findings

Circular emotion embeddings improve interpretability and robustness.

The approach underperforms in high-dimensional and fine-grained classification tasks.

Trade-offs exist between geometric interpretability and classification performance.

Abstract

Psychological research has long utilized circumplex models to structure emotions, placing similar emotions adjacently and opposing ones diagonally. Although frequently used to interpret deep learning representations, these models are rarely directly incorporated into the representation learning of language models, leaving their geometric validity unexplored. This paper proposes a method to induce circular emotion representations within language model embeddings via contrastive learning on a hypersphere. We show that while this circular alignment offers superior interpretability and robustness against dimensionality reduction, it underperforms compared to conventional designs in high-dimensional settings and fine-grained classification. Our findings elucidate the trade-offs involved in applying psychological circumplex models to deep learning architectures.