Sarcasm Identification in text with deep learning models and Glove word embedding

Abstract

In social media, sarcasm is frequently found to convey a negative opinion employing positive or exaggerated positive terms. Sarcasm is necessary for and advantageous for many Natural Language Processing (NLP) algorithms. Unless explicitly constructed to compensate for it, sarcasm can readily fool sentiment analysis tools. Sarcasm abounds in viewer stuff, such as Facebook postings and Tweets. It’s quite difficult to spot sarcasm without a complete awareness of the event, the wider problem, and the context. Memory-based network models, notably Long Short Term Networks (LSTM), Bidirectional Long-Short Memory (Bi-LSTM), and Convolution Neural Network (CNN) algorithms, can identify sarcastic remarks in a corpus. We assess the efficacy of competing Deep Learning algorithms for text sarcasm identification using the SarcasmV2 corpus. Based on the results, we can infer that for sarcasm detection, the Bidirectional Long Short Term Memory Network (Bi-LSTM) model provides the best performance. The suggested approach involving deep networks is consistent towards many traditional approaches for sarcasm detection, and based on specific standard performance indicators, the current model outperforms these approaches.