Behavioral Study of Large Language Models Based on the Influence of Emotions

In recent years, with the rapid advancement of artificial intelligence and deep learning technologies, large language models (LLMs) have demonstrated remarkable performance across various natural language processing tasks and are increasingly being applied in numerous contexts. Emotion, as an individual's integrated response to external events and internal stimuli, manifests not only as a psychological experience but also as physiological reactions, playing a crucial role in human cognition, behavior, and decision-making. Given that the design of neural networks is inspired by brain function, we explored whether and how emotional factors can influence the performance of LLMs. We investigated the impact of emotions on LLMs by introducing specially designed emotional prompts. We constructed prompts that encompass both positive and negative emotions, each with three distinct intensity levels. After embedding these emotional prompts into the inputs of LLMs, we systematically evaluated model performance across 24 instruction-induced tasks. Additionally, we conducted an in-depth analysis of the model's internal dynamics, including attention matrices and gradient information, to reveal how emotional stimuli affect the model's reasoning process. Our study elucidates the effects of emotions on LLMs, laying the foundation for future research on emotionally intelligent artificial intelligence.