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

“Large language models (LLMs) have achieved significant performance in many fields, such as reasoning, language understanding, and math problem-solving, and are regarded as an important step to artificial general intelligence (AGI). However, the sensitivity of LLMs to prompts remain a major bottleneck for their daily adoption. In this paper, we take inspiration from psychology and propose EmotionPrompt to explore emotional intelligence to enhance the performance of LLMs. Our EmotionPrompt operates on a remarkably straightforward principle: the incorporation of emotional stimulus into prompts. Experimental results demonstrate that our EmotionPrompt, using the same single prompt templates, significantly outperforms the original prompt and Zero-shot-CoT in both zero-shot and few-shot settings on eight tasks with diverse models: ChatGPT, Vicuna-13b, Bloom, and Flan-T5-large. Furthermore, Emotion- Prompt was observed to improve both the truthfulness and informativeness. We believe that EmotionPrompt heralds a novel avenue for exploring interdisciplinary knowledge for interaction between humans and LLMs.”

“We take the first step by applying psychological knowledge for LLMs enhancement. Previous studies in psychology showed that adding emotional stimulus to humans, which is related to expectancy, confidence, and social influence, can bring positive effects. Such examples widely exist in the real world, such as enhancing students’ success in education [Miltiadou and Savenye, 2003] and health promotion [Bandura, 1998] by using encouraging and positive words. Inspired by such psychology phenomenon, we propose EmotionPrompt, which is a simple yet effective approach for LLMs enhancement. Specifically, we design 11 sentences of emotional stimulus for LLMs, which can be simply added to original prompts and demonstrate this improvement. These emotional stimuli are simple psychological phrases that come after the original prompts. For instance, Fig. 1 shows an example of using one emotional stimulus, “This is very important to my career” at the end of the original prompts to enhance the performance of LLMs.”

“Positive words make more contributions. In our designed emotional stimulus, some positive words play a more important role, such as “confidence”, “sure”, “success” and “achievement”. Based on this finding, we summarize positive words’ contributions and their total contributions to the final result on eight tasks. As shown in Fig. 3, the contributions of positive words pass 50% on four tasks, even approach 70% on two tasks.”