Multilabel Emotion Detection from Bangla Text Using BiGRU and CNN-BiLSTM

Bangla is a popular language spoken by millions of people worldwide. These people use Bangla to convey their emotions. In addition, many non-natives but enthusiastic people in Bangla also want to interpret or communicate these emotions using the translator service. It would be nice to have a way out of the text to detect these feelings. Aiming this target, we have introduced two approaches to identify the emotion from a Bangla text. Taking deep learning as a cornerstone, we have developed two models to predict the affection of Bangla text. These two-prediction model has accounted for six identical emotion concerned labels to predict. To accomplish this, we relied on the Bidirectional Gated Recurrent Unit called BiGRU and the combination of the most classical deep learning algorithm, Convolutional Neural Network, and Bidirectional Long Short-Term Memory named CNN-BiLSTM. After exploiting these two algorithms, we calculated accuracy, precision, recall, and F1-score to evaluate their output. Since the dataset was assembled using a Google translator, those built models display promising performance for any Google Translator's translated extensive Bangla text.