HIMA-Net: humor prediction by self-attention based on key information related to humor

Humor is a high-level semantic emotion that can only be understood at a stage when the human mind has developed. Humor detection is a challenging task in the field of natural language processing. In this paper, we focus on the characteristics of humor from the way it is generated and propose the Humor Important Message Attention Net (HIMA-Net): a self-attention network based on the key messages related to humor. Results show that HIMA-Net outperforms the traditional models on three datasets (Headline, Pun, Short Jokes), and further analysis demonstrates the effectiveness of the proposed model.