Multilabel emotion detection from Bangla text containing emoji

Abstract

Emotion encompasses a wide range of human feelings. People express their emotions using a combination of facial expressions, body language, words, pictures, videos, texts, etc. Text messages accompanied by emojis and emoticons are commonly utilized in modern longdistance communication through social media platforms and A single line comment can convey multiple emotions. Sentiment analysis and emotion recognition are crucial problems in the field of natural language processing (NLP) and have extensive applications in several domains such as artificial intelligence, e-commerce, healthcare, education, human-computer interaction (HCI), business, and politics. Despite Bangla being the 7th most widely spoken language globally, it is classified as a low-resource language due to the scarcity of digital resources and linguistic tools. Authors have used several methods such as machine learning, deep learning, rule-based approaches, keyword-based approaches and hybrid techniques to extract emotions from text and emojis. In this study, we have used a scraping algorithm for collecting Bangla text data containing emojis from social media and construct an annotated corpus based on the scraped data. We have developed a comprehensive framework for recognizing multilabel emotions from text data that contains emojis. In this study, hybrid models and conventional machine learning techniques have been used for multilabel emotion recognition. Support Vector Machine outperforms other methods for individual emotion classes with 70.4% accuracy.