Emotion Classification in a Resource Constrained Language Using Transformer-based Approach

Although research on emotion classification has significantly progressed in\nhigh-resource languages, it is still infancy for resource-constrained languages\nlike Bengali. However, unavailability of necessary language processing tools\nand deficiency of benchmark corpora makes the emotion classification task in\nBengali more challenging and complicated. This work proposes a\ntransformer-based technique to classify the Bengali text into one of the six\nbasic emotions: anger, fear, disgust, sadness, joy, and surprise. A Bengali\nemotion corpus consists of 6243 texts is developed for the classification task.\nExperimentation carried out using various machine learning (LR, RF, MNB, SVM),\ndeep neural networks (CNN, BiLSTM, CNN+BiLSTM) and transformer (Bangla-BERT,\nm-BERT, XLM-R) based approaches. Experimental outcomes indicate that XLM-R\noutdoes all other techniques by achieving the highest weighted $f_1$-score of\n$69.73\\%$ on the test data. The dataset is publicly available at\nhttps://github.com/omar-sharif03/NAACL-SRW-2021.\n