How are You Really Doing? Dig into the Wheel of Emotions with Large Language Models

This study explores the application of Large Language Models (LLMs) for Emotion Analysis on social media, focusing on predicting 80 diverse emotions. After comprehensive evaluations, fine-tuned Llama 70B emerges as the top-performing model across datasets, showcasing its capacity to capture a broad spectrum of emotions effectively. Comparisons with traditional baselines highlight the superior performance of LLMs, emphasizing their effectiveness in understanding and predicting emotions in social media posts. In this study, we have also investigated prompting strategies and we discussed the minimal impact that different prompts have on LLMs performance. The study’s implications span social media analysis, sentiment analysis, and customer service, with potential future research exploring novel prompts, assessing model transferability, and addressing biases in emotion prediction.