Multi-task RoBERTa for sarcasm detection via Sentiment Incongruence Auto-Labeling. Focal Loss + WeightedRandomSampler + LIME explainability. F1-Macro 0.977, AUC 0.997. Zero-shot cross-domain eval on ABSA & Amazon. IEEE paper.
nlp data-science natural-language-processing deep-learning sentiment-analysis text-classification transformers pytorch data-analytics cross-domain multi-task-learning sarcasm-detection focal-loss roberta huggingface lime-explainability implicit-sentiment
-
Updated
Apr 22, 2026 - Jupyter Notebook