Publication

Abstract : This paper describes the submission of the Amrita_CEN_NLP team to the shared task on Dravidian- CodeMix-FIRE2021. The dataset used in this task is CodeMix text associated with the context of social media. It’s most common to notice the comments under Youtube videos, Facebook posts in the CodeMix. In this task, we implemented three different Deep learning-based architectures: Deep Neural Network (DNN), Bidirectional-Long Short Term Memory network (Bi-LSTM), and finally, Convolution Neural network (CNN) combined with a Long Short Term Memory network (LSTM) for predicting various sentiments associated with the Dravidian CodeMix languages(Malayalam, Tamil, Kannada). The data given by organizers is highly imbalanced to handle this issue weightage given to each class weight based on their distribution over data. Our experiments reveal that CNN combined with LSTM, DNN with one hidden layer performs best for Malayalam linguistics and, the BiLSTM layer suits the classification of Tamil and Kannada corpus. After inferring the results obtained on performed experiments, we submitted the results.