Home // BRAININFO 2024, The Ninth International Conference on Neuroscience and Cognitive Brain Information // View article

Emotional Recognition and Classification Using Large Language Models

Authors:
Clement Leung
Zhifei Xu

Keywords: image emotion recognition, large language model, zero-shot, ChatGPT4.

Abstract:
Many tasks, particularly safety-critical ones, require the associated human performers to be in the right emotional states. Correct emotion state recognition frequently becomes an important concern and mainstream methods often use Pretrained Language Models (PLMs) as the backbone to incorporate emotional information. The latest Large Language Models (LLMs), such as ChatGPT have demonstrated strong capabilities in various natural language processing tasks. However, existing research on ChatGPT zero-shot has received insufficient evaluation of the performance of image emotion recognition and analysis. In this paper, we study emotion classification and prediction based on positive and negative emotional states, and evaluate the emotion recognition capabilities of ChatGPT4 focusing primarily on images. We empirically analyze the impact of labeled emotion recognition and interpretability of different datasets. Experimental results show that, while ChatGPT is able to make some useful predictions of emotions based on images, there is still a substantial gap in prediction results and accuracy. Qualitative analysis shows its potential compared to state-of-theart methods, but it also suffers from limitations in robustness and accurate inferences.

Pages: 4 to 10

Copyright: Copyright (c) IARIA, 2024

Publication date: March 10, 2024

Published in: conference

ISSN: 2519-8653

ISBN: 978-1-68558-127-5

Location: Athens, Greece

Dates: from March 10, 2024 to March 14, 2024