Multimodality Deep Feature-Based Stress Classification Using Mita Algorithm

Abstract

Clients generate enormous volumes of data massively and dynamically these days, thanks to the growing variety of Web technologies. In this context, emotional analysis seemed to be a key technique for automating the extraction of understanding from user-generated material. So far, sentiment classification on data from social media has been limited to a single modality, such as text or picture. However, easily available multimodal information, such as images and other types of texts, as a group can assist in more precisely anticipating thoughts. Deep learning has recently demonstrated exceptional performance in the field of emotion and is regarded as the advanced approach. In this research, we present a Multimodal Image and Text with Attention (MITA) algorithm for classifying the sentiment of social media tweets by combining the Backpropagation Long Short-Term Memory with Conditional Random Field (BiLSTM-CRF) and Fusion awareness models for both text and image. Each character's context text is modeled by the BiLSTM layer. The BiLSTM layer's hidden states are handled at the CRF layer to improve sequential labeling with the help of neighboring labels. On the other hand, the image is addressed with a basic structure of Convolutional neural Networks (CNNs) including three perspectives such as spatial, a channel with squeezing, and excitation at various levels. Our suggested effort yielded encouraging results and enhanced sentiment classification accuracy.