Large Language Models for Emotion Evolution Prediction

In numerous tasks, especially those of critical safety importance, it is essential that the human participants maintain appropriate emotional states. Recognizing these emotional states accurately has become a key focus, with mainstream approaches typically leveraging Pre-trained Language Models (PLMs) to integrate emotional insights. Recent advancements in Large Language Models (LLMs), like ChatGPT, have shown impressive proficiency in a range of natural language processing applications. Yet, the exploration into ChatGPT’s zero-shot capabilities in the realm of image-based emotion recognition and analysis has been notably under-examined. In our research, we study the classification and prediction of emotions, distinguishing between positive and negative states, and we critically assess the ability of ChatGPT4 to recognize emotions from images, putting a spotlight on its potential in this area. The experimental results show that ChatGPT4 can effectively predict the evolution of emotion, and the accuracy of the evolution of positive and negative emotions is more than expected. However, ChatGPT4’s prediction of specific negative emotions has gaps.