Enhanced Sarcasm and Irony Detection in Social Media Using LSTM and Global Average Pooling
M. Ramprasath, A. V. Kalpana, R. Seetharaman, G. Elangovan, T. Nadana Ravishankar, M. Vijay Anand · 2024
Complex linguistic elements like sarcasm and irony are used in social media communication and require a nuanced understanding of tone and context. This paper presents a state-of-the-art approach to natural language processing that can detect and understand irony and sarcasm in social media writing by combining lengthy short-term memory with the Global Average Pooling Model. We tackle this problem by developing new deep-learning models that can pick up on the nuances, interdependencies, and mood swings that are characteristic of sardonic or sarcastic remarks. To go a step further, we investigate how combining multimodal data like pictures and emojis with transfer learning from large language models could improve the accuracy of irony and sarcasm identification. We demonstrate the effectiveness of our suggested models through rigorous evaluation using benchmark datasets and real-world social media content. The results of this study are very important since they provide a huge step forward in understanding complex linguistic subtleties in online conversations. The results of this study showed that LSTM+GAP outperformed other methods with an accuracy of 96.71% and have important consequences for our knowledge of online community motion, opinion mining, and sentimental analysis.