Sarcasm Detection using Bert Embeddings and Hybrid LSTM-CNN Architecture

Irony, a prevalent literary device, often appears in online writings and dialogues, complicating sentiment analysis. Sentiment analysis seeks to ascertain the emotional tone of a digital test-positive, negative, or neutral, rendering it an effective instrument for comprehending individuals' ideas and feelings in settings such as social media, reviews, and emails. This research concentrates on identifying irony in text via the use of breakthroughs in Natural Language Processing. We present a hybrid model that integrates Long Short-Term Memory (LSTM) networks with Convolutional Neural Networks (CNN) to assess, store, and identify irony in phrases. Furthermore, we use BERT (Bidirectional Encoder Representations from Transformers) for pre-training, augmenting the model's capacity to comprehend contextual linkages. Our results indicate that sarcasm detection markedly improves with the presence of contextual similarity across phases, underscoring the importance of linguistic and emotional subtleties. The model's performance, with an accuracy of 0.7750, highlights its applicability in sentiment analysis across various datasets. These observations have ramifications for improving automated text comprehension in areas such as consumer feedback, social media analysis, and conversational artificial intelligence.