Towards Enhanced Identification of Emotion from Resource-Constrained Language through a novel Multilingual BERT Approach

Emotion identification from text has recently gained attention due to its versatile ability analyze human-machine interaction. This work focuses on detecting emotions textual data. Languages, like English, Chinese, and German are widely used for classification, however, limited research is done resource-poor oriental languages. Roman Urdu (RU) a resource-constrained language extensively across Asia. predicting RU text. For this, dataset collected different social media domains based Paul Ekman's theory it annotated with six basic emotions, i.e., happy, surprise, angry, sad, fear, disgusting. Dense word embedding representations of languages adopted that utilize existing pre-trained models. BERT additionally fine-tuned the classification task. The proposed approach compared baseline machine learning deep algorithms. Additionally, comparison current also performed approaches same Based empirical evaluation, performs better than state-of-the-art an average accuracy 91%.