# EmotionPrompt: Leveraging Psychology for Large Language Models Enhancement via Emotional Stimulus

**Authors:** Cheng Li, Jindong Wang, Kaijie Zhu, Yixuan Zhang, Wenxin Hou, Jianxun Lian, Xing Xie  
**arXiv:** [2307.11760](https://arxiv.org/abs/2307.11760)  
**Date:** 2023-07-14  
**Summary page:** https://www.summarizepaper.com/en/paper/2307.11760v1/

## Key points

- EmotionPrompt is a novel approach inspired by psychology that incorporates emotional stimulus into prompts to enhance Large Language Model (LLM) performance.
- Experimental results show that EmotionPrompt significantly outperforms original zero-shot prompts and Zero-shot-CoT on 8 diverse tasks across various LLM models.
- EmotionPrompt not only enhances task performance but also improves truthfulness and informativeness.
- The contributions of this research are threefold: introducing EmotionPrompt as an innovative method rooted in psychological science, showcasing substantial performance improvement through extensive experiments, and providing insightful analysis drawing inspiration from both AI and social sciences.

## Summary

In this paper, the authors address the challenge of prompt sensitivity in Large Language Models (LLMs) by proposing EmotionPrompt, a novel approach inspired by psychology that incorporates emotional stimulus into prompts to enhance LLM performance. The study aims to leverage emotional intelligence to improve tasks such as determining whether a movie review is positive or negative. Experimental results demonstrate that EmotionPrompt significantly outperforms original zero-shot prompts and Zero-shot-CoT on 8 diverse tasks across models like ChatGPT, Vicuna-13b, Bloom, and T5. Notably, EmotionPrompt not only enhances task performance but also improves truthfulness and informativeness. The contributions of this research are threefold: firstly, it introduces EmotionPrompt as an innovative method rooted in psychological science to enhance LLMs; secondly, extensive experiments showcase the substantial performance improvement brought about by EmotionPrompt in various tasks; thirdly, insightful analysis delves into the rationale behind EmotionPrompt, drawing inspiration from both AI and social sciences. In recent years, there has been a surge in the development of large language models (LLMs), which have shown impressive performance on various natural language processing tasks. However, these models often struggle with prompt sensitivity - their ability to accurately interpret and respond to different prompts. To address this challenge, the authors propose a novel approach called EmotionPrompt. This method draws inspiration from psychology and incorporates emotional stimulus into prompts to enhance LLM performance. The study leverages emotional intelligence - a set of abilities related to processing emotional information - to improve tasks such as sentiment analysis. These abilities include perceiving emotions, using emotions for cognitive activities, understanding emotions,and managing emotions in oneself and others. Experimental results demonstrate that EmotionPrompt significantly outperforms original zero-shot prompts and Zero-shot-CoT on 8 diverse tasks across models like ChatGPT, Vicuna-13b, Bloom, and T5. This highlights the effectiveness of EmotionPrompt in improving LLM performance. The contributions of this research are threefold: firstly, it introduces EmotionPrompt as an innovative method rooted in psychological science to enhance LLMs; secondly, extensive experiments showcase the substantial performance improvement brought about by EmotionPrompt in various tasks; thirdly, insightful analysis delves into the rationale behind EmotionPrompt, drawing inspiration from both AI and social sciences. This paves the way for interdisciplinary exploration in enhancing human-LLM interaction through emotional intelligence.