Exploring the Capabilities of LLMs in Emoji Sentiment Analysis and Recommendation for Tweets

Emojis have become an essential element of online communication, enriching textual messages with nuanced emotional and contextual cues. Large Language Models (LLMs) have been widely recognized for their strong performance across a broad range of language understanding and reasoning tasks. The capabilities of different LLMs in understanding and using emojis in diverse tasks are still not well-studied. This study explores two important tasks by using tweets: (1) the ability of different LLMs to analyze emoji sentiments; (2) the ability of different LLMs to recommend appropriate emojis. We conduct the experiments using the 50 most frequently used emojis from 2019 to 2022 and a large tweet dataset. We explore the broader potential of LLMs by conducting a comparative analysis using alternative LLM models beyond ChatGPT. The results show that, (1) for emoji sentiment analysis, LLMs can capture emoji inherent sentiment, the sentiment changes in text aligns with the inherent sentiment of the emoji in many cases, and our study with diverse LLMs draws conclusions with higher confidence than the studies only using one LLM such as ChatGPT; (2) for emoji recommendation, although there were still challenges regarding recommendation accuracy, some LLMs have the potential to perform emoji recommendations and can take community context into account.