Semantic Representation is Superior to Syntactic Representation for Emotion Classification Using Graph Neural Networks
Nitish Vashisht · 2023
The vast volume of user-generated material available on popular social media platforms may be mined for insights about people's emotional states. There are several advantages to this, including learning how people react to the news and events. Short text length and the requirement to identify many emotions (a multi-label classification problem) make social media post emotion categorization challenging. Deep neural networks, including Convolutional neural network designs and recurrent neural network models, have been used in the majority of past work on emotion categorization. However, in order to categorize a broad variety of emotions, none of these systems have attempted to gather semantic and syntax information from text. In this research, we suggest using semantics and syntax aware graphs of attention to categorize emotions from unlabeled text. To finish off the multimodal emotion identification, a graph convolutional neural network (GCNN) is deployed. Extensive trials on two actual data sets, IEMOCAP and MELD, show that the GANNs model achieves greater average accuracy along with f1 scores than the current models for multimodal emotion detection, particularly for emotions like “happiness” and “anger.” As a result, the model's emotional intelligence may be improved with the help of dependent syntactic analysis and a self-attention mechanism.