Sarcasm Detection Revisited: Reproducing and Comparing Top-Performing Techniques on a Standardized Dataset

Sarcasm detection remains a challenging task in natural language processing (NLP) due to the subtle and context-dependent nature of sarcastic expressions. This paper revisits sarcasm detection by reproducing and comparing top-performing machine learning, deep learning, and transformer-based techniques on a standardized dataset. We structure our work using the CRISP-DM methodology, encompassing data collection from various sources, cleaning, and transformation. We implement several baseline machine learning models, including Naive Bayes and logistic regression, and explore deep learning models to capture more complex patterns in the data. Our key contribution is the development and fine-tuning of a hybrid BERT-BiLSTM model, which leverages BERT's powerful contextual embeddings and BiLSTM's ability to capture sequential dependencies, resulting in enhanced sarcasm detection performance.