Year: 2024
Type: conference-paper
Abstract: In recent years, large language models have become a top trend in the field of natural language processing. A series of language models, represented by GPT-4, trained with massive amounts of text data, have provided a completely new experience different from traditional agents. However, when trying to imitate human rational logic, large language models seem to lack the ability to model human emotional representations in complex and uncertain environments, especially the uncertainty caused by emotional fluctuations. Based on this, we propose ECNet, a method for modeling human emotions in large language models. First, we fine-tune the LLM with specific data. Then, through three steps—Emotion Perception, Emotion Confirmation, and Emotion Reinforcement—we complete the modeling of human complex emotions and uncertain fluctuations. We invited several participants to subjectively evaluate the effectiveness of ECNet. The results show that our method greatly enhances the personalization of LLMs and achieves the integration of emotion and logic. Our method makes a significant contribution to emotion perception and expression in natural language processing.
Authors: Xinshan Qin, Conglin Ma, Yan Hu, Maochao Tian
Institution: North China University of Technology
Language: English
Topic: Emotion and Mood Recognition
Subfield: Experimental and Cognitive Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: gold
OpenAlex ID: W4411088622