A Label Extension Schema for Improved Text Emotion Classification

Abstract

Due to the subjectiveness and fuzziness of emotions in texts, researchers have been aware that it is ubiquitous to observe multiple emotions in a sentence, and the one-hot label approach is not informative enough in emotion-relevant text classification tasks. Therefore, to facilitate the classification task, recent works focus on generating and employing a coarse-grained emotion distribution, which is based on coarse-grained labels provided by the underlying dataset. Although such methods can alleviate the problem of overfitting and improve robustness, they may cause inter-class confusion between similar emotion categories and introduce undesirable noise during training. Meanwhile, current studies neglect the fine-grained emotions associated with these coarse-grained labels. To address the issue caused by utilizing a coarse-grained distribution, we propose in this paper a general and novel emotion label extension method based on fine-grained emotions. Specifically, we first identify a mapping function between coarse-grained emotions and fine-grained emotion concepts, and extend the original label space with specific fine-grained emotions. Then, we generate a fine-grained emotion distribution by employing a rule-based method, and utilize it as a model constraint to incorporate the dependencies among fine-grained emotions to predict the original coarse-grained emotion labels. We conduct extensive experiments to demonstrate the effectiveness of our proposed label extension method. The results indicate that our proposed method can produce notable improvements over baseline models on the applied datasets.