Explainability in Irony Detection

Ege Berk Buyukbas, Adnan Harun Dogan, Asli Umay Öztürk, Pinar Karagoz

Abstract: Irony detection is a text analysis problem aiming to detect ironic content. The methods in the literature are mostly for English text. In this paper, we focus on irony detection in Turkish and we analyze the explainability of neural models using Shapley Additive Explanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME). The analysis is conducted on a set of annotated sample sentences.