ExaAEC: A New Multi-label Emotion Classification Corpus in Arabic Tweets
Saeed Sarbazi-Azad, Ahmad Akbari, Mohsen Khazeni · 2021
Nowadays the usage of social media tools by individuals to share their emotions and opinions in the form of short text messages is increasing. Emotion has a significant role in people’s interactions. Detecting Emotion specially from short texts in social media, has become a challenging problem in natural language processing. Despite recent deep learning successes in a variety of fields of NLP, previous studies on low resource languages mostly suffer from scarcity of data. In this research, we are going to introduce a new dataset of emotion recognition on Arabic tweets called ExaAEC which contains about 20000 tweets annotated by six native Arab taggers. The aforementioned data is labeled in ten classes inspired from the Plutchik emotion model. To classify the emotions, Embedding from Language Models (ELMO) and Long Short Term Memory (LSTM) neural networks are applied. According to the results, the F1-score in the experiments was 0.65 which indicates, our data and model can perform adequately in Emotion Detection. The corpus is publicly available here1.