Result: A Deep Learning-Based Framework for Sentiment and Emotion Classification of Social Media Messages During Pandemic Periods.

Twitter is an emerging social media platform with more than 700 billion users worldwide. It is an active platform for spreading information, sharing opinions, expressing emotions, etc. It helps technical researchers understand the mental ability of specific situations during pandemic periods such as COVID-19 and FLU. Recently, COVID-19 has been a severe pandemic that has affected the world's economic rate, and lots of people have lost their lives. To understand the severity of COVID-19, the government has announced a lockdown so people worldwide can share their emotions through social media platforms. This initiated many researchers to research sentiment analysis and emotion classification in Twitter messages to help the government sectors take necessary and accurate actions. However, they failed to provide accurate results. In this paper, deep learning-based sentiment and emotion classification (SENTI-EMO) is proposed to provide accurate results. Initially, the pre-processing of Twitter messages, such as Noise removal, correction, tokenization and normalization, is done to improve the classification accuracy. In the noise removal phase, the stop word technique is used to learn about the context of tweets to enhance further classification results. After pre-processing, sentiment classification is done by using attention-based GRU, which contains several layers to extract high-level features; in the attention layer, only essential features are considered to reduce redundancy, which enhances the accuracy of sentiment classification. The extracted features classify the sentiments as strongly positive, positive, negative, negative and neutral. The classified sentiments are clustered by using the spatiotemporal optics (STO) algorithm, which forms clusters based on Twitter ID, geographical location, and time stamp. Then, by using the Multi-class CatBoost algorithm, the clustered sentiments are classified into different emotions based on various topics and locations, such as fear, anger, trust, joy and sadness, to reduce overfitting, latency, and class imbalance issues and enhance emotion classification accuracy. Finally, all the reports are generated for the government sector, from there they use soft actor-critic (SAC) algorithm to accurately take actions based on the reports which enhance the detection accuracy. The process is implemented in the Python 3 tool. The proposed work performs better than existing works in terms of validation metrics such as Accuracy (avg-92.04%), Precision (avg-93.2%), Recall (avg-92.58%), F-score (avg-91.8%) and computation time (avg-74.8 ms). [ABSTRACT FROM AUTHOR]

Copyright of Journal of Circuits, Systems & Computers is the property of World Scientific Publishing Company and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)