Unlocking Cross-Lingual Sentiment Analysis through Emoji Interpretation: A Multimodal Generative AI Approach

As the world is getting interconnected, the need for effective universal sentiment indicators is more critical than ever before. A universal medium of sentiment analysis across diverse languages can reduce global communication challenges. However, the existing work has not yet fully leveraged the multimodal capabilities of ChatGPT to explore the capacity of emoji to serve as a universal sentiment indicator in a cross-lingual context. This analysis aims to illuminate the capacity of emojis to serve as a global medium of sentiment across diverse languages. Our study investigated how various emoji representations analyzed through large language models (LLMs) convey the same emotion as text with emojis. First, our analysis showed that emojis separated from the original text convey the same sentiment as the text itself with an accuracy of 81.43%. Second, we found that most of the available emojis are neutral. Lastly, we observed that the accuracy of sentiment conveyed by emojis gradually increased as the number of emojis grew in tweets. The research indicates that emojis have significant potential to enhance sentiment analysis, offering cost-effective solutions for cross-lingual applications using LLMs.