Sentiment and Emotion in Social Media COVID-19 Conversations: SAB-LSTM Approach

Abstract

Sentiment and Emotion detection in social media conversations remains a challenge and analyzing the people emotion emerged as an important task in this unprecedented time of COVID-19. People sentiment and emotions are affected by lockdowns, social distancing, travel, work-from-home, wearing mask, reading social media posting. Most of them are feeling sad, anger, depressed, and some of them are neutral and happy. The most recent Sentiment Analysis (SA) researches are done using Twitter dataset (short-text) and rule-based (sentiment lexicon) approach, the outcome of these SA models' results is not showing the consistent prediction of people sentiment about COVID-19. To mitigate and overcome limitations of lexicon approach, processing unstructured social media long text posting, getting context based sentiment score, model overfitting, performance problems of sentiment models, authors' proposed and built a novel multi-class SA model using extension of Bidirectional LSTM (SAB-LSTM) with additional layers. In this experiment SAB-LSTM model has been used to process long text of social media posting, news articles text dataset. Experiment result showed, SAB-LSTM model performance is better than traditional LSTM and BLSTM. Compared SAB-LSTM performance metric of Precision, Recall, F1 Score and sentiment score with traditional LSTM and BLSTM. For this experiment collected COVID-19 related dataset from various social media sources such as Twitter, Facebook, YouTube, News articles blogs and collected data from friends and family.