An Approach for Emotion Detection Using a Hybrid Vision Sequence-Based Convolutional Neural Networks (Vi-CNN) Model

Emotion represents the significant reflection or indicator of human mental states, biological effects, and physiological statements, and it also plays a vital role in human interpersonal communication and decision-making. People convey their emotions naturally via everyday interactive communication due to the increasing advancement of social media platforms. Nowadays, detecting emotions from extensive textual data helps to provide expressive information for understanding the behavior of humans. Meanwhile, most prior techniques used for emotion detection are insufficient in providing promising results from long-term contextual information. Thus, it motivates introducing a hybrid deep learning model, Vision sequence-based Convolutional Neural Network (Vi-CNN), from text data to detect emotion. The Bidirectional Encoder Representation from Transformer (BERT) transforms the input texts into tokens. Then, extracting the most suitable and appropriate features from the text tokens is executed. Following this, the Vi-CNN model significantly detects emotions from the extracted features and classifies the recognized emotions. Furthermore, the results obtained by Vi-CNN are compared with those of other prevailing schemes. The experimental results highlight that the Vi-CNN attained promising results with maximum recall of 94.765%, precision of 92.988%, and F-measure of 93.867%.