Sarcasm Detection for Sentiment Analysis using Deep Learning Enhancement

Sarcasm is a verbal irony where the intended message contrasts with the literal meaning, often used to mock or express contempt. Sarcasm is context-dependent and requires nuanced verbal clues, tone, and context, which are challenging to capture in text-based analysis. Traditional sentiment analysis models are insufficient due to the complexity of sarcasm and cultural and individual disparities. However, new developments in deep learning, particularly transformer architecture, show promise for improving sarcasm detection accuracy. In this paper, We propose a novel method combining transformer augmentation with hybrid optimized deep learning techniques such as LSTM, BiLSTM, CNN, and pre-trained transformer-based BERT to detect complex sarcasm from social platforms. Experiments conducted on benchmark datasets reveal that our system works well, with notable gains in sarcasm detection accuracy over baseline techniques. We achieved 96% accuracy, 97% precision, 93% recall, and 95% F1-Score by the proposed model. Our model introduces a reliable and accurate way to find sarcasm in different text types by combining transformer upgrades and optimized hybrid DL techniques to benefit social media monitoring, customer feedback analysis, and opinion mining, where understanding genuine emotion is crucial.