Fine grained irony classification through transfer learning approach

Nowadays irony appears to be pervasive in all social media discussion forums and chats, offering further obstacles sentiment analysis efforts. The aim of the present research work is detect its types English tweets We employed a new system for detection tweets, we propose distilled bidirectional encoder representations from transformers (DistilBERT)light transformer model based on (BERT) architecture, this strengthened by use design long-short term memory (Bi-LSTM) network configuration minimizes data preprocessing tasks proposed tests SemEval-2018 task 3, 3,834 samples were provided. Experiment results show has achieved precision 81% not class 66% class, recall 77% 72% irony, F1 score 79% 69% since researchers have come up with binary classification model, study extended our multiclass irony. It significant will serve as foundation future different tweets.