Emojis are used in text to convey emotion, express individual style, and reflect personality. Existing emoji prediction frameworks depend on textual embeddings while overlooking features such as user behavior, and conversation context. This paper provides a comprehensive analysis of the impact of emotion, usage patterns, and personality traits on emoji prediction. Additionally, it presents a personalized emoji prediction framework that integrates emotion, usage patterns and user personality. Experimental results show that the proposed framework outperforms the baseline text-only model, achieving gains of up to 2% in accuracy and 3% in precision. Furthermore, we introduce a semantic evaluation approach that groups emojis based on their semantic similarity. Using this evaluation approach, the personalized model continues to outperform the baseline text-only model, confirming its effectiveness in providing more semantically relevant emojis.
Gaafar et al. (Thu,) studied this question.