Multi-Modal Sarcasm Detection in Weibo

Abstract

Irony is a writing technique with a sense of irony, and the original meaning it expresses is just the opposite of what can be understood literally. However, the existing sarcasm detection is mostly based on text analysis and seldom pays attention to the influence of emoji pictures on the expression of text emotion. This article focuses on multi-modal satire detection for tweets composed of text and emoticons on Weibo. Firstly, image features and expression features are extracted, and then expression features and Bi-LSTM network are used to extract text features. Finally, the three modal features are fused for prediction. We created a Weibo-based multi-modal irony detection data set, and the evaluation results on the data set proved the effectiveness of our proposed model.