Impact analysis of emotion in figurative language

Abstract

Due to the implicit traits embedded in tweets, handling figurative languages appear as the most trending topics in computational linguistics. While recognition of a single language is hard to capture, differentiating several languages at once is the most challenging task. To achieve this purpose, we employ a set of emotion-based features in order to individuate between humor, irony, sarcasm, satire and true. We use eight basic emotions excerpted from EmoLex supplement with tweets polarity. We apply these features in two datasets: balanced dataset (collected using hashtag-based approach) and class-imbalanced dataset (collected from streaming tweets). As a result, the model not only outperform a word-based baseline but also handle both balanced and class-imbalanced datasets in multi-figurative language detection.