Emotion Recognition in Social Network Texts Based on A Multilingual Architecture

Languages are used by people to describe and categorize their emotional experiences and perspectives. For many applications, it is crucial to apply techniques like machine learning in social network texts to identify emotions. Most of these technologies now in use only detect a small number of emotion categories such as anger, happiness, sadness and so on, they do not distinguish more fine-grained levels of emotions. Additionally, they frequently concentrate on modeling the relationships between various emotions, ignoring the emotional semantic relations between different languages. Therefore, in this paper, we improve the Recognition of Emotion by utilizing a Multilingual architecture that combines machine Translation and Attention mechanism, enabling one language to provide additional emotional information for another language (REMTA). The experimental results on a fine-grained emotion dataset labeled with 28 categories show a performance improvement compared with other models, demonstrating the efficacy of our architecture.