The limitations of irony detection in Dutch social media

Abstract In this paper, we explore the feasibility of irony detection in Dutch social media. To end, investigate both transformer models with embedding representations, as well traditional machine learning classifiers extensive feature sets. Our feature-based methodology implements a variety information sources including lexical, semantic, syntactic, sentiment features, two new data-driven features to model common sense. Based on patterns syntactic structure tweets, aim presence contrasting sentiments, phenomenon that is known be indicative verbal and sarcasm. Feature selection, voting ensemble techniques were implemented enhance classification performance. The final systems reach F1-scores up 0.79, which are promising results for task difficult detection. Besides quantitative analysis, paper also describes thorough qualitative analysis system output. Although lexical cues appear very important express irony, our revealed need more advanced modeling common-sense knowledge detect subtle examples irony.