Label-Aware Hyperbolic Embeddings for Fine-grained Emotion Classification

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 10947–10958 July 9-14, 2023 ©2023 Association for Computational Linguistics Label-Aware Hyperbolic Embeddings for Fine-grained Emotion Classification Chih-Yao Chen1, Tun-Min Hung2, Yi-Li Hsu2, Lun-Wei Ku2 1UNC Chapel Hill, 2Institute of Information Science, Academia Sinica 1cychen@cs.unc.edu 2{allenhung,yili.hsu,lwku}@iis.sinica.edu.tw Abstract Fine-grained emotion classification (FEC) is a challenging task. Specifically, FEC needs to handle subtle nuance between labels, which can be complex and confusing. Most existing models only address text classification problem in the euclidean space, which we believe may not be the optimal solution as labels of close semantic (e.g., afraid and terrified) may not be differentiated in such space, which harms the performance. In this paper, we propose Hy- pEmo, a novel framework that can integrate hyperbolic embeddings to improve the FEC task. First, we learn label embeddings in the hy- perbolic space to better capture their hierarchi- cal structure, and then our model projects con- textualized representations to the hyperbolic space to compute the distance between sam- ples and labels. Experimental results show that incorporating such distance to weight cross en- tropy loss substantially improves the perfor- mance with significantly higher efficiency. We evaluate our proposed model on two bench- mark datasets and found 4.8% relative improve- ment compared to the previous state of the art with 43.2% fewer parameters and 76.9% less training time. Code is available at https: //github.com/dinobby/HypEmo. 1 Introduction Fine-grained classification is a challenging yet im- portant task that involves differentiating subtle dis- tinctions in a label set. For instance, in image classification, classifying cars, planes, and other vehicles is coarse-grained classification, whereas distinguishing models from cars is fine-grained classification. In NLP, sentiment analysis, which attempts to classify positive/negative sentiments, is an example of coarse-grained text classification. Human emotions, however, exhibit more complex- ity. For example, the six type of basic emotion (Ek- man, 1999) include happiness, sadness, fear, dis- gust, anger, and surprise, and show finer distinc- tions of positive and negative classes. Moreover, Figure 1: Properties of hyperbolic space. The hyper- bolic distance between two points grows dramatically as they come close to the border, converse to the Eu- clidean distance. Moreover, the hyperbolic distance is an analog of tree distance, which is d(B, C) ≈ d(B, A) + d(A, C). complex interactions exists in human emotion as different type of emotions can have subtle differ- ences, for instance ashamed and guilty. This makes fine-grained emotion classification (FEC) challeng- ing not only because of the increased number of classes, but also because of the increased similarity between classes. For instance, the current finest emotion classification datasets contain up to 27 and 32 classes of emotion (Rashkin et al., 2019; Dem- szky et al., 2020), respectively. Classes such as furious and angry in these fine-grained datasets are far more difficult to differentiate than happy and sad. However, detecting fine-grained emotion is useful in a variety of situations. For example, in a dialog generation system, understanding the user’s fine-grained emotion could …

Content truncated to 500 words