ESRM: A Model for Deep Semantic Analysis of Emojis based on Prompt Learning
Jun Zhang, Qinru Li, Hongli Deng, Rong Zeng · 2023
Emojis, as a way of users' emotional expression, are widely used in social scenes such as comments and communications. Due to the use of irony, borrowing, and other linguistic techniques, the shallow semantics of emojis do not reflect the real intentions expressed by users. This makes traditional models based on the shallow semantics of emoji misjudge the relationship between emoji and context, and thus make wrong predictions and classifications. To address this problem, this paper proposes an Emoji Semantic Recognition Model (ESRM) based on prompt learning. This model uses prompt learning to set up targeted prompt templates and integrate the deeper meaning features of emojis for social information recovery. Through experiments, it is demonstrated that the ESRM proposed in this paper can effectively semantically analyze social messages containing implicit expressions. The incorporation of the deep meanings of emojis makes ESRM maximize the accuracy of hidden content recovery by about 10.75%.