Enhancing Sentiment Analysis using BERT-Hybrid Model for Detection of Irony and Sarcasm in Code-Mixed Social Media

Abstract

Sarcasm is a common verbal irony that can be difficult to apply in code-mixed social media, as people fluently switch between languages. The difficulty of sarcasm detection in these kinds of language environments is the subject of this study. Conventional models find it difficult to capture the complex interactions between different languages, slang terms, and cultural quirks. As a result, we suggest an Hybrid model based on BERT that efficiently detects sarcasm by utilizing the contextual knowledge of BERT embeddings. Combining different BERT versions, the Hybrid model performs better at detecting and interpreting sarcasm in code-mixed social media. Our method greatly advances the state of natural language processing through extensive experimentation, especially when it comes to managing the intricacies of multilingual and multicultural online communication. Keywords: Sarcasm detection, code-mixing, BERT Hybrid model, linguistic complexity.