Sarcasm Detection in News Headlines Using Evidential Deep Learning-Based LSTM and GRU

Md. Shamsul Rayhan Chy, Md. Shamsul Rahat Chy, Mohammad Rakibul Hasan Mahin, Mohammad Muhibur Rahman, Md Sabbir Hossain, Annajiat Alim Rasel

Abstract: Sarcasm has become quite inter-related with the day to day life of all. In news robust sarcasm is often used to grab the attention of the viewers. This research aims to detect sarcasm using Evidential deep learning. This technique uses uncertainty estimations for identifying the sentiments from news headlines dataset. Also, LSTM and GRU have been used with Evidential deep learning approach. The purpose of using LSTM is that it can classify texts from headlines in order to analysis the sentiments. Moreover, we have used GRU which is an recurrent neural networks (RNN) and it effectively models sequential data. The architecture of the GRU network is ideally suited for identifying dependencies and extended contextual relationships within news headings. Overall, our proposed model uses Evidential deep learning based LSTM and GRU to identify the sentiments of robust sarcasms from news headlines.