SarcasmDet at Sarcasm Detection Task 2021 in Arabic using AraBERT Pretrained Model

摘要

This paper presents one of the top five winning solutions for the Shared Task on Sarcasm and Sentiment Detection in Arabic (sub-task 1 Sarcasm Detection). The goal of the sub-task is to identify whether a tweet is sarcastic or not. Our solution has been developed using ensemble technique with AraBERT pre-trained model. This paper describes the architecture of the submitted solution in the shared task. It also provides in detail the experiments and the hyperparameters tuning that lead to this outperforming result. Besides, the paper discusses and analyzes the results by comparing all the models that we have trained or tested to build a robust model in a table design. Our model is ranked fifth out of 27 teams with an F1-score of 0.5989 of the sarcastic class. It is worth mentioning that our model achieved the highest accuracy score of 0.7830 in this competition.