Fusing Emoji Emotion Distribution for Multi-Label Emotion Classification
Su-Fen Chen, Ye Liu, Xueqiang Zeng · 2024
With the rise of Internet-based social media, emoji has become a widely used image text for users in daily communication due to its feature of expressing emotions quickly and accurately in a graphical way. Existing studies have shown that considering emoji information in text-based emotion recognition models plays an important role in improving model performance. At present, most emotion recognition models considering emoji information adopt the word embedding models to learn the emoji representations, and the emoji vectors obtained lack direct correlation with the target emotion, and the emoji representations contain less emotion recognition information. To address the above problems, this paper constructs an emotion distribution vector for emoji directly associated with the target emotion through soft labels and combines the emoji emotion distribution information with text semantic information based on the pre-training model, and proposes the Emoji emotion distribution Information Fusion for multi-label Emotion Recognition method (EIFER). Based on the classical binary cross-entropy loss function, EIFER method models the correlation between emotional labels by introducing label-correlation aware loss, so as to improve the multi-label emotion recognition performance of the model. The model structure of the EIFER method is composed of a semantic information module, an emoji information module and a multi-loss function prediction module, and the model is trained in an end-to-end way. Compared experiment results of emotion prediction on the SemEval2018 English dataset have shown that the proposed EIFER method has better performance than the existing emotion recognition methods.