Emojis—those expressive yellow faces and diverse icons that many of us use in texts and emails—offer a fun and quick way for people to share sentiments, convey emotions, and establish aspects of their identity that can’t easily be communicated through plain text.
But for some, including researchers in the University of Maryland’s Computational Linguistics and Information Processing (CLIP) Lab, those smiling faces and fire emojis are not just conversational spice, but instead represent powerful data points that can improve machine learning and natural language processing (NLP) models.
The UMD team is exploring how emojis can advance the development of more robust algorithms able to adapt to the nuanced ways people express themselves online. This new knowledge is essential for improving technology used for language translation software, sentiment analysis—analyzing text to determine if the emotional tone of the message is positive, negative or neutral—and user activity prediction, widely used to track online shopping.
For their latest project, the UMD team investigated ChatGPT’s ability to annotate text and emojis that are used to build machine learning and NLP models.
Typically, researchers rely on human crowdsourcing to annotate emojis to better understand their sentiments, usage intentions, and semantic meanings, says Yuhang Zhou, a fifth-year doctoral student in the College of Information.
Not only is this method time consuming, Zhou explains, but it also isn’t scalable, and the human annotators’ differences in demographics, cultural background, or personal experience can lead to subjective misunderstandings of emoji meanings.
In a paper that will be presented in later this year at the Association for the Advancement of Artificial Intelligence’s International Conference on Web and Social Media in Denmark, Zhou—who is the lead author—argues that ChatGPT can reliably serve as a potential alternative to human annotators in emoji research.
The researchers—including faculty and graduate students from UMD’s College of Information and Department of Computer Science, and the University of Arizona’s School of Information—challenged ChatGPT to myriad tasks involving emojis, such as assessing the positive, negative or neutral sentiment of a tweet based on whether emojis were present.
The varied outputs showed that ChatGPT considers the meaning of the emoji when assessing sentiment, says co-author Wei Ai, an assistant professor in the College of Information with an appointment in the University of Maryland Institute for Advanced Computer Studies.
The research team also proved that ChatGPT understands cultural context around certain emojis. For example, it accurately interprets the goat emoji 🐐as “the greatest of all time.”
Another key finding was the platform’s ability to detect irony in English, Chinese and Arabic depending on whether emojis were included. The researchers believe that this accurately illustrates ChatGPT’s nuanced understanding of emojis’ role in conveying irony.
Overall, the team concluded that in addition to ChatGPT offering a precise understanding of emoji functionality and intention, it can also be reliable tool for social media users to decipher the semantics of unfamiliar emojis, therefore enhancing the clarity and transparency of communication online.
—Story by Maria Herd, UMIACS communications group