A Comprehensive Emotion Recognition from Microblogs Using Deep Learning
Md. Masudur Rahman, M. A. H. Akhand, Md Abdus Samad Kamal, Tetsuya Shimamura · 2021
Emotion is one of the fundamental features of human beings. With the advancement of the internet, social media platforms (e.g., Facebook, Twitter, Instagram) became an integral part of social interaction in daily life. Microblog is one of the popular platforms for sharing opinions and has become an important source for emotion research and analysis. Human holds different emotion states (e.g., Happy, Sad, Angry), but most existing studies are concerned with a few selected emotion categories. This study aims to develop a Comprehensive Emotion Recognition from Microblogs (CERM) system intelligently managing available microblog contents (texts, symbols) to recognize all the standard emotional categories: Anger, Fear, Joy, Love, Sad, Surprise, Disgust, and Neutral. As the use of emoticons (graphical symbols) is being popular in microblogs, the main attraction of the study is the management of both emoticon and text for better performance. From Twitter, various microblogs (i.e., tweets) samples are collected, processed, and used to classify all the standard emotion categories. The most popular deep learning methods, including Convolutional Neural Network and Long Short-Term Memory, are employed for emotion classification, and the best suited deep learning model is identified from the experiments. The proposed CERM system is confirmed to outperform related existing methods in comparison based on recognition accuracy.